In October, I posted all my answers to Canada’s AI Strategy Survey. But I have questions of my own, and this is one of them.

Nearly everyone recognizes that the current level of interest and investment in AI will not be sustained indefinitely. How will the public and policy conversations change once AI falls out of the news cycle? Or, worse yet, once some high profile companies become financial victims of the hype cycle?

Let’s imagine that we were tasked with setting internet policy in 2003, after the dotcom crash and the public launch of MySpace - but before the founding of Facebook and the mobile web. An entire industry lost momentum and there was little interest in shaping its future at the policy level. But what a window of opportunity it could have been!

Such a window of opportunity may soon open up for current AI technologies. How will we use it?

웃https://okhuman.com/WH77OQ

I gave each of the 26 questions serious thought and published all my answers in a series of posts.

Given what was asked -and just as importantly - what wasn’t, I have three of my own questions for the task force. Here is the first.

1. What is AI?

There is a running joke that AI is what we call new algorithms that don’t yet work reliably - or that hasn’t found use cases beyond demos. Old but established algorithms are not exciting, despite being powerful and useful tools. In the 2010s, image classification using deep learning experienced tremendous progress. Exciting and bold predictions were made about how it would replace radiologists and truck drivers. This has not happened, but these algorithms have found widespread adoption in innumerable use cases. The excitement has waned despite such algorithms being an integral part of any modem computer vision toolkit.

Machine learning was AI in the 2010s after its AlexNet moment, where it matched humans in a broad image recognition benchmark. But these methods are not part of the current AI conversation - a conversation purely focused on generative AI methods. Will Canada’s new AI strategy take into account machine learning and other older methods that are seeing less progress, less attention but can still have tremendous impact? What about even older and simpler methods, such as support vector machines? Looking forward, how will the strategy consider today’s’ bleeding edge methods once they become routine tools and new, unproven methods are in the spotlight instead?

This is the final post in “Responses to the Canada AI Survey” - a series containing my responses to the eight themes of the Canadian government’s public consultation on artificial intelligence.

View all posts in this series

After working my way through the survey, reading what many others wrote and attending some roundtable discussions, here are some critical questions that I believe the task force will need to develop answers for.

What is AI?

There is a running joke that AI is what we call new algorithms that don’t yet work reliably - or that haven’t found use cases beyond demos. Such algorithms are not exciting despite being powerful and useful tools. In the 2010s, image classification using deep learning (a branch of machine learning which itself is a type of artificial intelligence) experienced tremendous progress. Exciting and bold predictions were made about how it would replace radiologists and truck drivers. This has not happened, but these algorithms have found widespread adoption in innumerable use cases. The excitement has waned despite such algorithms being an integral part of any modern computer vision toolkit.

Machine learning was AI in the 2010s after its AlexNet moment, where it matched humans in a broad image recognition benchmark. But these methods are not part of the current AI conversation - a conversation purely focused on generative AI methods. Will Canada’s new AI strategy take into account machine learning and other older methods that are seeing less progress, less attention but still have tremendous impact? What about even older and simpler methods, such as support vector machines? Looking forward, how will the strategy consider today’s bleeding edge methods once they become routine tools and new, unproven methods are in the spotlight instead?

What happens when the AI hype fizzles?

Nearly everyone admits that the current level of interest and investment in AI will not be sustained indefinitely. How will the public and policy conversations change once AI falls out of the news cycle? Or worse yet, once some high profile companies become financial victims of this hype cycle. Let’s imagine that we were tasked with setting internet policy in 2003, after the dotcom crash and the public launch of MySpace - but before Facebook and the advent of mobile Web. How will we handle this window of opportunity with LLMs?

If AI lives on the global marketplace of the internet, how could we ever control it?

The internet is a global marketplace, in particular because a great deal of its currency in the consumer space is based on attention and advertising instead of money. AI in the consumer space, but also in the cloud-enabled enterprise space is also distributed on this global marketplace. What are the realistic expectations in shaping AI on the web if it can freely enter and exit Canadian borders? Does our ideal of a free and open internet come under pressure?

How do we shape the use of AI if incentives drive behaviour and outcomes?

Many of our greatest technological disasters, including those which are still ongoing, are unsurprising as consequences of existing incentives. Is there any goodwill to change such incentives given AI’s potential? Many are drawing attention to what this means in the extreme case - our capitalistic society may destabilized by rapid adoption of AI due to displacement of labour and other effects. Which incentives at the individual, corporate, political and societal level can be subject to modification?

Where does technology not considered AI fit in?

More than 90% of doctors in Ontario still use fax machines. It’s painful to spell it out, but the technology to replace the fax machine has existed for decades now - and none of it required AI. Will Canada’s AI Strategy ignore opportunities to make advances in areas that also need updated policy and incentives but that don’t require actual AI?

Moving Forward

The Canada AI Survey represents a valuable exercise in stakeholder consultation, but it’s only the beginning of the conversation. The 30 day sprint was short and hopefully comes from a recognition that we need to move quickly. I hope that we can implement much needed change with an eye to the past, and continue adapting with an eye to the future. Our current ChatGPT moment is too narrow a scope to dictate far-reaching policy decisions.

This is part 8 of “Responses to the Canada AI Survey” - a series containing my responses to the eight themes of the Canadian government’s public consultation on artificial intelligence.

View all posts in this series

Theme 8: Security of the Canadian Infrastructure and Capacity

Q1: What are the emerging security risks associated with AI, and how can Canada proactively mitigate future threats?

Every application of AI that is deployed becomes a new attack surface. This is particularly challenging when this AI is non-deterministic. Doubly so if the software was written with AI tools.

However, we cannot guard against tools and content we don’t create. Phishing and impersonation attempts are already clear examples, where the level of sophistication has increased with the help of AI.

Q2: How can Canada strengthen cybersecurity and safeguard critical infrastructure, data and models in the age of AI?

Canada should further strengthen software development practices in the face of the following which each create new ways for security holes to appear: 1) AI software development tooling 2) Deployed AI models and functions these models perform 3) data these models ingest to guard against threats such as prompt injection.

We need strong, human-coded guardrails for each use case and deployment. Cybersecurity professionals are sufficiently spooked by the dangers of AI. We need to extend current practices in both preventing and addressing security flaws.

Q3: Where can AI better position Canada’s protection and defence? What will be required to have a strong AI defensive posture?

There has been a large, but mostly fruitless effort, to detect AI. However, as we move from assuming content is real to assuming it is fake until proven otherwise, it starts making sense to start authenticating real content instead. Certain sectors should have policies and interoperable systems for content provenance and authenticity. These can be for strictly internal use, but also can benefit the wider public, particularly in the media landscape.

This is part 7 of “Responses to the Canada AI Survey” - a series containing my responses to the eight themes of the Canadian government’s public consultation on artificial intelligence.

View all posts in this series

Theme 7: Building and Enabling Infrastructure

Q1: Which infrastructure gaps (compute, data, connectivity) are holding back AI innovation in Canada, and what is stopping Canadian firms from building sovereign infrastructure to address them?

I don’t think compute is holding back innovation in Canada. It is a commodity which is easily accessed across borders from private enterprise such as AWS. The price of GPU-enabled compute has actually fallen for late-generation chips. There are signs of a disconnect between rental prices and purchase pricing which, for a depreciating asset, is problematic. This is a serious consideration for any public infrastructure build-out.

Data has, for a long time, been digital gold. There has been a large effort to access data in the healthcare technology space, particularly medical images after AlexNet in 2012. However, with the advent of natural language processing and now LLMs, unstructured patient data is also sought. This is data that is not available on the web. Canada should guard all its medical data from US firms if it seeks an advantage. We are behind in this respect. We actually use other country’s data in some instances for non-AI guidance, such as bone-mineral density screening. This is an ongoing effort with many barriers remaining.

All verticals have valuable data housed in silos. Data owned by Canadian firms (and ostensibly coming from Canadians or Canadian activities and processes) should be leveraged to a Canadian advantage. It may make sense for industries to cooperate and the government could have a role in establishing neutral parties to aid this cooperation, both from legal and technical (e.g. data lakes) standpoints.

Connectivity varies widely across the country, especially in rural areas. We should address this immediately. Rural connectivity projects often take many years after commitments are made to achieve the actual connections. This is too slow.

Q2: How can we ensure equitable access to AI infrastructure across regions, sectors and users (researchers, start-ups, SMEs)?

Access is currently equitable in the sense that it’s mediated by internet connectivity and money for anything in the current “LLM stack”. Outside of this stack (research), we need to follow the lead of these researchers and their academic institutions. SMEs may want to access such specialized stacks if some capability arises that isn’t already on the open market. At this point, I think it is a question of market forces - the cost of expanding this stack vs. the revenues it could create. So long as it falls short of a renewed promise of an imminent AGI which requires immediate, massive scaling, as is the case with current LLMs.

Q3: How much sovereign AI compute capacity will we need for our security and growth, and in what formats?

It’s long been clear that in a globalized world, Canada is unable to onshore everything, especially as so much focus is on resource extraction and automotive manufacturing.

At minimum, Canada should have sovereign Content Delivery Networks that are located and wholly owned by Canadians. The same for data centers.

However, it is likely unreasonable to manufacture our own chips or even replicate the data center build out happening elsewhere.

There is a potential scenario where the buildout south of the border has happened too soon and too quickly. This could be an advantage, where these resources could be accessed for cheaper than it would have cost to build them. We could even purchase them at a steep discount and power them with our own green electricity.

Lastly, I’m going to challenge the assumption that more compute equals more security. If more compute is needed because of more use and more use means more applications and each application is a new, novel attack surface, perhaps more compute equals less security.

This is part 6 of “Responses to the Canada AI Survey” - a series containing my responses to the eight themes of the Canadian government’s public consultation on artificial intelligence.

View all posts in this series

Theme 6: Education and Skills

Q1: What skills are required for a modern, digital economy, and how can Canada best support their development and deployment in the workforce?

Labour predictions are impossible. Policy makers need to humbly recognize that nearly every time an occupation has been predicted to become irrelevant, the timeline and/or the technological impact was too optimistic. This leads to a near term shortage and the adoption of less desirable secondary solutions. We also need to recognize that some predictions, especially those espoused by an industry, are self-fulfilling.

The question seeks an answer in the modern, digital economy, but we’ve long had a shortage of trades in many areas in Canada. I propose there is an appreciable percentage of white collar knowledge workers that may actually derive immense satisfaction from learning a trade. We could perhaps learn something from the Swiss vocational system where people participate in both apprenticeships and classroom-based learning and an engineer is often someone who was initially trained as a machinist.

In more vague terms, we should encourage education paths that train students in ways to think and approach problems in a more general sense. These are usually education programs that have historically had unclear career prospects (other than academia). Physics and philosophy are two that come to mind.

In the strictest sense, “AI-focused” careers will be a very small percentage of the workforce. It would be the equivalent of seeking to be ready for “cloud-focused” careers 15 years ago.

There is a large group of occupations where AI cannot replace humans or that we may want to resist this replacement. We should direct a great deal of effort to attract both the existing and future workforce to these roles, especially those that have a positive impact on human quality of life.

Canada is facing such a moment right now. The situation with Canada Post is a result of technological shift and automation in the face of inflexible historical norms.

Q2: How can we enhance AI literacy in Canada, including awareness of AI’s limitations and biases?

To begin, what do we mean by “AI literacy” will need to be defined. We leverage automobiles for transportation and also depend on them. Are all drivers “car literate” or only those that do their own oil changes or have read the owner’s manual? Or do we speak of the mechanics who can actually troubleshoot and resolve problems?

I don’t believe AI literacy is required from the public to achieve our goals. As technology has improved, we have become less “car literate” - fewer people change their own oil than ever before and oil level checks are done electronically by the car’s system. Nearly everyone uses the cloud in some way, understands the concept of it, but we haven’t done anything to increase “cloud literacy”, beyond the advertisement of cloud providers.

This is not to say that there are not any real dangers in terms of limitations, biases, misinformation, fake content, etc. But the guardrails and the safeguards are required of the systems that perform the functions, the systems that distribute them and the processes that use the outputs.

In the end, literacy is a very individual trait - it’s unfair and misguided to pin this requirement on the public. Our politicians and policy makers, however, should absolutely be AI literate, but they have so far often fallen short of it. Let’s start there.

Q3: What can Canada do to ensure equitable access to AI literacy across regions, demographics and socioeconomic groups?

There are (much more) fundamental things lacking equitable access across Canada. AI literacy is knowledge and knowledge travels freely on the internet (if one can cut through the noise and stay out of echo chambers).

Let’s start with equitable internet access - and more importantly - clear water and access to affordable housing and food.

This is part 5 of “Responses to the Canada AI Survey” - a series containing my responses to the eight themes of the Canadian government’s public consultation on artificial intelligence.

View all posts in this series

Theme 5: Building Safe AI Systems and Strengthening Public Trust in AI

Q1: How can Canada build public trust in AI technologies while addressing the risks they present? What are the most important things to do to build confidence?

The most important thing is to do the “right thing” - even if the right thing is admitting to uncertainty or even mistakes. External communication from researchers, government and industry needs to be proactive. The media will need to somehow report on potential and known issues while avoiding the typical catastrophizing or exaggeration incentivized by their business model. Falling behind in communication after incidents will quickly spoil trust.

Grounding public trust in AI on the public’s AI literacy would be extremely unwise. It would also be unfair to the public. Politicians and policy makers have yet to show such AI literacy themselves. Confidence will be built over time by avoiding incidents - and minimizing the politicisation of use cases, especially for those in the public sector.

Q2: What frameworks, standards, regulations and norms are needed to ensure AI products in Canada are trustworthy and responsibly deployed?

Most professional sectors have existing frameworks and oversight bodies to address safety and ethics (medicine, construction, transportation, etc). We should not duplicate their efforts but instead let them take the lead and listen to their calls for additional legislation where needed. Blanket legislation will have unpredictable and unintended consequences otherwise.

Some domains will remain without adequate oversight. I submit that these domains are already fraught with integrity and ethical concerns which have long needed to be addressed, well before our current AI moment. These should be a priority. Oversight should be created that is independent of technology and which can manage the principal incentives giving rise to the problems.

Q3: How can Canada proactively engage citizens and businesses to promote responsible AI use and trust in its governance? Who is best placed to lead which efforts that fuel trust?

Incentives are key. Proactive responsible AI use means recognizing and tackling misaligned incentives. Trust is earned, it is an outcome of other actions. Any PR, advertisement or body with a primary purpose of “fuelling” trust immediately has the opposite effect as soon as there is reason to doubt that trust - especially in the age of social media. This question feels misguided by positioning trust as a primary goal. Just do the right thing.

This is part 4 of “Responses to the Canada AI Survey” - a series containing my responses to the eight themes of the Canadian government’s public consultation on artificial intelligence.

View all posts in this series

Theme 4: Scaling Canadian Champions and Attracting Investments

Q1: How does Canada get to more and stronger AI industrial champions? What supports would make our champions own the podium?

Canada’s industrial AI champions should be strong in Canada - while being set up for rapid expansion broadly outside of it. But, they can’t necessarily be identified ahead of time - and they may not be AI companies in an obvious or marketable way. Therefore, I’m unfortunately proposing this question comes from a flawed perspective. Champions need to arise naturally and whether they are AI champions also falls out naturally of their business. We should not choose them, especially if doing so could come at a disadvantage to other Canadian prospects.

Q2: What changes to Canada’s landscape of business incentives would accelerate sustainable scaling of AI ventures?

Incentives should align to adoption and procurement of Canadian-built solutions - even if some uncertainty exists. There is currently an aversion to adoption of solutions from small companies and it’s widely recognized that this needs to change, across sectors, but the public sector should lead the way. This allows early stage companies to iterate on product-market fit which must be achieved before capital can accelerate growth.

Q3: How can we best support AI companies to remain rooted in Canada while growing strength in global markets?

The two goals of keeping Canadian companies rooted in Canada while growing in global markets are not diametrically opposed. We also have to accept that there may be companies who offer solutions that are perfect for the Canadian market but do not find the same success globally. These are success stories too. I think these should be treated as separate goals, so proposed by some of the other survey questions. Also, the solutions should apply to non-AI companies.

Q4: What lessons can we learn from countries that are successful at investment attraction in AI and tech, both from domestic sources and from foreign capital?

Competitive environments seem to attract the most investment. Global investors are not looking for a sure thing, they are looking for the potential of a big win. This includes the recruitment of talent, which requires talent mobility. For example, talent mobility is extremely high in California, aided by a prohibition on non-compete clauses. This also encourages the founding of new ventures in similar fields. Employment law in Canada is much murkier and varies province to province in this respect.

There is also, in my opinion, a problem of optics currently when it comes to early stage ventures. Why would global capital invest in Canadian firms when Canadians are not freely adopting their solutions AND Canadian investors are not investing in them? The Canadian ecosystem has in part responded to this by creating direct bridges to ecosystems like Silicon Valley from which to draw investments. In their own response, Canadian capital experiences a sort of loss aversion after it becomes clear they have missed out. While they may create new groups and funds to address this issue, they must proactively convince founders that they can find their first capital here.

We want the foreign investors to be the ones who feel like they are missing out.

This is part 3 of “Responses to the Canada AI Survey” - a series containing my responses to the eight themes of the Canadian government’s public consultation on artificial intelligence.

View all posts in this series

Theme 3: Commercialization of AI

Q1: What needs to be put in place so Canada can grow globally competitive AI companies while retaining ownership, IP and economic sovereignty?

If Canadian companies can have Canadian customers with whom to iterate and develop the product to achieve financial sustainability within Canada, they will be in a much better position to be globally competitive. Currently, most early stage Canadian start-ups, particularly those within regulated environments, look for traction outside of Canada because it is both easier with a better payoff in terms of both adoption and investments - most of which will then be foreign.

Q2: What changes to the Canadian business enabling environment are needed to unlock AI commercialization?

It is well known that early stage startups in Canada have an easier time raising money south of the border than in their own backyard. This needs to change. I’m unsure on how to increase the appetite of Canadian investors on this front. Someone should ask them why they wait for a company to gain traction in Silicon Valley before answering their emails.

Sectorally, opportunities should be identified by experts - from that sector. Places where money, time and other resources are wasted and where AI could play a big role. Perhaps tax credits could be granted for money saved by adopting technology - particularly if this capital is then redirected for growth or R&D.

Q3: How can Canada better connect AI research with commercialization to meet strategic business needs?

Academia is, except for rare instances, quite removed from immediate business needs. Today’s AI moment, powered by LLMs, genAI and computer vision are ready to be applied and are based on the academic work of the prior decade.

Domain experts who know everything about their domain but perhaps nothing about AI have a huge role to play. They are best positioned to identify problems that actually need solving (the nails) while they can team up with AI enablement experts to identify which hammer to wield. Many lessons can be learned from the application of CNNs to radiology, where workflow considerations were paramount, as they will be in other domains.

This is part 2 of “Responses to the Canada AI Survey” - a series containing my responses to the eight themes of the Canadian government’s public consultation on artificial intelligence.

View all posts in this series

Theme 2: Accelerating AI Adoption by Industry and Government

Q1: Where is the greatest potential for impactful AI adoption in Canada? How can we ensure those sectors with the greatest opportunity can take advantage?

In industry and professional domains, AI adoption is wholly workflow-dependent. Many AI projects are already failing because of a failure to recognize this. Each vertical requires its own domain experts that are either well versed in AI capabilities or who can collaborate with those who are to identify areas of impactful AI adoptions.

I can speak to healthcare, where it is easy to imagine the great impact to finances, efficiency and care outcomes. However, there are many impactful technologies which don’t require AI that face steep barriers to adoption already in our healthcare systems. In other words, we must remove those barriers that prevent the adoption of standard technologies as well.

Q2: What are the key barriers to AI adoption, and how can government and industry work together to accelerate responsible uptake?

Any existing barrier to technology adoption will also slow down the adoption of AI. Many have probably already been identified by leaders in specific sectors. We should start there.

It is well known that consensus paralysis prevents many innovation projects from ever launching. A single no can stop an idea completely. We need a culture shift in this regard, especially in the public sector. Perhaps there should be a branded effort behind this effort - Elbows Up for Innovation.

Old software that stands in the way of innovation should be replaced. There will be a temptation to replace older, but still recent software with a version that is AI-enabled. But I believe there is more potential in using AI to replace software that is very old. The kind of software that only a handful of people are still touching to maintain. AI tooling should be used to rebuild it in a cost effective way, and AI should be used to interface it with other systems. Temporary stitching of such systems can now be cost effective with AI.

Multiple stakeholders, liability of rogue projects (CIOs not knowing what’s going on), accelerate update by incentivizing replacing old, old software where AI can be the glue that allows it to work with new software, or the tool that allows a replacement to be cost efficient. A lot of code is touched by the very few, let alone by someone equipped with AI. Government should procure home grown AI solutions.

Q3: How will we know if Canada is meaningfully engaging with and adopting AI? What are the best measures of success?

This is a dangerous question - we all know of metrics that became hindrances the moment they become targets.

Any metrics should be based on sector-specific measures of efficiency and satisfaction. Anything based on compute, tokens, api calls etc. will be unaligned with efficiencies. Further, there are sometimes non-AI solutions (or at least non-LLM) that should be adopted. After all, simpler but powerful statistical methods such as support vector machines were once considered AI.

There should also be metrics that relate to people as they become educated and employed in fields touched by AI - including how many leave.

This is part 1 of “Responses to the Canada AI Survey” - a series containing my responses to the eight themes of the Canadian government’s public consultation on artificial intelligence.

View all posts in this series

Theme 1: Research and Talent

Q1: How does Canada retain and grow its AI research edge? What are the promising areas that Canada should lean in on, where it can lead the world?

It’s crucial to recognize that our current AI moment is based solely on the scaled implementation of Generative Pretrained Transformers (GPT). The technology itself is nearly 10 years old. GPTs are an extension of a series of machine learning techniques introduced through the 1980s, 1990s, and 2010s.

These methods would not have looked promising to many people at the time. They also would have had to compete with other research avenues that, though abandoned now, may have held promise at the time. For this reason, it may not be possible to identify today’s promising areas.

Canada can maintain its research edge by exploring the non-promising and non-obvious at the research stage while looking far ahead to changing capabilities that could break current assumptions about technical feasibility (such as massively scaling highly parallelized computation in the case of GPTs). For example, symbolic AI approaches offer an alternate paradigm to approaches based solely on machine learning and neural nets and could be complementary if combined or scaled in a yet to be discovered way.

Canada will not grow its research edge by competing with private industry to scale GPTs. GPTs have flaws and inherent performance limitations that are intrinsic to their structure and training. They are not the path to artificial general intelligence. Canada therefore needs to explore many other avenues to find those which may offer a stepwise increase in capabilities.

Q2: How can Canada strengthen coordination across academia, industry, government and defence to accelerate impactful AI research?

Canada should double down on the programs that are working well, tweak the ones that are not working as well as they should - all while removing as much bureaucracy as possible from them. In particular, post-doctoral researchers are highly trained, creative individuals and usually underpaid for the amount of long term value they can create. While this position was often a stepping stone to further a career in academia, industrial post-doctoral fellowships should be seen as an entryway into a lucrative technical leadership career in any sector.

Q3: What conditions are needed to ensure Canadian AI research remains globally competitive and ethically grounded?

Core ethical principles are core to our society and evolve slowly, even as technology changes rapidly. We need to equip existing oversight bodies instead of creating new ones. Due to the speed of progress, we should centralize thought leadership and debate for the most impactful areas of concern.

The slow pace of legislation may negatively impact both global competitiveness by creating uncertainty, while failing to protect people with real consequences for bad actors. For this reason, legislation should be proactive while being non-specific. It should not seek to ban or restrict specific technologies but instead focus on outcomes.

Research ethics have existed for a long time and institutions have dedicated offices and staff for their oversight. We need to make it easy for these offices to adapt to a changing landscape while cutting through the noise. I don’t think new entities are needed. I believe it would be worth centralizing thought leadership on the topic, in a Canadian cultural and legal context, for that purpose.

Q4: What efforts are needed to attract, develop and retain top AI talent across research, industry and the public sector?

Canada should continue to position itself as an attractive place for AI talent, and particularly for AI talent working at Canadian companies and institutions. The fluid situation in the United States with respect to international students and H1B visas is an opportunity to further cement Canada’s reputation as a welcoming place for this talent. Canada could see a boost in reputation that outlives these restrictions south of the border.

The academia-to-industry pathway may be one of the best opportunities to retain top talent. It should be better developed with increasing industrial partnerships and remuneration. A culture of exploration should be encouraged, where post-docs can cross disciplines at this juncture instead of only being offered to double-down on their areas of expertise. I would even go so far as renaming the industrial post-doctoral fellowship to differentiate it from the academic track and capturing the fact that we are developing the innovation leaders of tomorrow.

This is part 12 of “101 Ways AI Can Go Wrong” - a series exploring the interaction of AI and human endeavor through the lens of the Crossfactors framework.

View all posts in this series or explore the Crossfactors Framework

Here’s a principle every genAI booster should take note of - the Collingridge Dilemma.

What is it

The Collingridge Dilemma describes how technologies in their early stages are easy to control, but nearly impossible to predict, while in later stages, as their consequences become clearer, the opportunity to control them has passed.

Why It Matters

The Collingridge Dilemma has become incredibly relevant in recent decades, as the internet and computers have allowed the proliferation and scaling of technologies at previously impossible rates. At the extreme, this dilemma is embodied by the fear of AGI take-off. However, long before this may happen, the self-replicating and self-modifying features of future AI deployments will continue to shorten the Window of Opportunity where technologies can be influenced in their formative phases.

Real-World Examples

An example of the Collingridge Dilemma that evolved from both our digital and physical spaces is the proliferation of ride-sharing platforms such as Uber and Lyft in urban settings.

The disruption of traditional taxi services was initially seen as a net positive by the public because of the combination of smart-phone based hailing and lower prices. Advantages were also seen on the supply side, as drivers could use their existing vehicles on their own schedule outside the regulatory oversight of traditional taxis.

As Uber and Left proliferated several serious issues started to become apparent:

• Exploitation of gig workers.
• Increased traffic congestion due to vehicles circling around waiting for a fare.
• Predatory pricing at peak times.
• Erosion of public transit.
• Increase of prices after traditional taxis were no longer around to compete and from venture capital was removed.
• Abuse of the platform by both drivers and riders.

The long term impact of ride-sharing services is still playing out now, through legal and regulatory battles, traditional taxis evolving with their own technology, local legislative changes, and labour disputes.

Key Dimensions

The Anticipatory Gap - We must recognize that the technology is nascent and that we are blind to the eventual outcomes.

Legislation - Legislation must recognize the above and anchor itself into fundamental human principles. It must allow for adaptive governance and domain-specific regulation, like in medicine.

Trial and error - The reach and stakes must be incremented through trial and error. This is what will allow us to learn from our mistakes before they are too widespread to fix.

Take-away

The Collingridge Dilemma explains why we need to build with humility, not hype.

Simply put - no. There are far too many details that require domain-specific knowledge for this to be possible.

Frameworks are extremely useful - and necessary. Amongst other things, they provide a variety of very specific lenses from which to view the safe implementation of AI.

But there’s plenty that can fall through a framework’s gaps. Think of idiosyncrasies within a specific application area. Such things are well known and understood by experts who operate in this domain, but by few outside of it.

Another thing frameworks will miss are unintended, higher order effects, which are often unpredictable, and surprising. Whats not surprising however is that they often include a human element.

They are published by industry groups, corporations, academic groups, governments, standard agencies and even notable individual voices in this space.

These frameworks come in different flavours. Responsible AI, AI Safety, AI deployment, AI implementation, AI enablement, AI Risk Management, AI governance…and more, I’m sure!

Additionally, many corporations are developing their own bespoke and private enterprise AI frameworks which are often inspired and adapted from other published frameworks. Ultimately, like any business function, there is no one size-fits-all solution.

Such frameworks take different approaches in compartmentalising the various functions that underpin the use of AI at the enterprise level. They vary in how inclusively they span all related functions. Lastly, they may bring a particular lens to their approach, such as a security, data or lifecycle centric approach.

These frameworks are all making one thing very clear. There are a lot of things to get right. This also means there are a lot of ways things can go wrong!