GUEST POST from Pete Foley
Some of the biggest questions of our age are whether AI will ultimately benefit or hurt us, and how big its’ effect will ultimately be.
And that of course is a problem with any big, disruptive technology. We want to anticipate how it will play out in the real world, but our forecasts are rarely very accurate, and all too often miss a lot of the more important outcomes. We often don’t anticipate it’s killer applications, how it will evolve or co-evolve with other emergent technologies, or predict all of the side effects and ‘off label’ uses that come with it. And the bigger the potential impact new tech has, and the broader the potential applications, the harder prediction becomes. The reality is that in virtually every case, it’s not until we set innovation free that we find its full impact, good, bad or indifferent.
Pandora’s Box
And that can of course be a sizable concern. We have to open Pandora’s Box in order to find out what is inside, but once open, it may not be possible to close it again. For AI, the potential scale of its impact makes this particularly risky. It also makes any meaningful regulation really difficult. We cannot regulate what we cannot accurately predict. And if we try we risk not only missing our target, but also creating unintended consequences, and distorting ‘innovation markets’ in unexpected, potentially negative ways.
So it’s not surprising there is a lot of discussion around what AI will or will not do. How will it effect jobs, the economy, security, mental health. Will it ‘pull’ a Skynet, turn rogue and destroy humanity? Will it simply replace human critical thinking to the point where it rules us by default? Or will it ultimately fizzle out to some degree, and become a tool in a society that looks a lot like today, rather than revolutionizing it?
I don’t even begin to claim to predict the future with any accuracy, for all of the reasons mentioned above. But as a way to illustrate how complex an issue this is, I’d like to discuss a few less talked about scenarios.
1. Less obvious issues: Obviously AI comes with potential for enormous benefits and commensurate problems. It’s likely to trigger an arms race between ‘good’ and ‘bad’ applications, and that of itself will likely be a moving target. An obvious, oft discussed potential issue is of course the ‘Terminator Scenario’ mentioned above. That’s not completely far fetched, especially with recent developments in AI self preservation and scheming that I’ll touch on later. But there are plenty of other potential, if less extreme pitfalls, many of which involve AI amplifying and empowering bad behavior by humans. The speed and agility AI hands to hackers, hostile governments, black-hats, terrorists and organized crime vastly enhanced capability for attacks on infrastructure, mass fraud or worse. And perhaps more concerning, there’s the potential for AI to democratize cyber crime, and make it accessible to a large number of ‘petty’ criminals who until now have lacked resources to engage in this area. And when the crime base expands, so does the victim base. Organizations or individuals who were too small to be targeted for ransomware when it took huge resources to create, will presumably become more attractive targets as AI allows similar code to be built in hours by people who possess limited coding skills.
And all of this of course adds another regulation challenge. The last thing we want to do is slow legitimate AI development via legislation, while giving free reign to illegitimate users, who presumably will be far less likely to follow regulations. If the arms race mentioned above occurs, the last thing we want to do is unintentionally tip the advantage to the bad guys!
Social Impacts
But AI also has the potential to be disruptive in more subtle ways. If the internet has taught us anything, it is that how the general public adopts technology, and how big tech monetizes matter a lot. But this is hard to predict. Some of the Internet’s biggest negative impacts have derived from largely unanticipated damage to our social fabric. We are still wrestling with its impact on social isolation, mental health, cognitive development and our vital implicit skill-set. To the last point, simply deferring mental tasks to phones and computers means some cognitive muscles lack exercise, and atrophy, while reduction in human to human interactions depreciate our emotion and social intelligence.
1. Cognitive Obesity The human brain evolved over tens of thousands, arguable millions of years (depending upon where in you start measuring our hominid history). But 99% of that evolution was characterized by slow change, and occurred in the context of limited resources, limited access to information, and relatively small social groups. Today, as the rate of technological innovation explodes, our environment is vastly different from the one our brain evolved to deal with. And that gap between us and our environment is widening rapidly, as the world is evolving far faster than our biology. Of course, as mentioned above, the nurture part of our cognitive development does change with changing context, so we do course correct to some degree, but our core DNA cannot, and that has consequences.
Take the current ‘obesity epidemic’. We evolved to leverage limited food resources, and to maximize opportunities to stock up calories when they occurred. But today, faced with near infinite availability of food, we struggle to control our scarcity instincts. As a society, we eat far too much, with all of the health issues that brings with it. Even when we are cognitively aware of the dangers of overeating, we find it difficult to resist our implicit instincts to gorge on more food than we need. The analogy to information is fairly obvious. The internet brought us near infinite access to information and ‘social connections’. We’ve already seen the negative impact this can have, contributing to societal polarization, loss of social skills, weakened emotional intelligence, isolation, mental health ‘epidemics’ and much more. It’s not hard to envisage these issues growing as AI increases the power of the internet, while also amplifying the seduction of virtual environments. Will we therefore see a cognitive obesity epidemic as our brain simply isn’t adapted to deal with near infinite resources? Instead of AI turning us all into hyper productive geniuses, will we simply gorge on less productive content, be it cat videos, porn or manipulative but appealing memes and misinformation? Instead of it acting as an intelligence enhancer, will it instead accelerate a dystopian Brave New World, where massive data centers gorge on our common natural resources primarily to create trivial entertainment?
2. Amplified Intelligence. Even in the unlikely event that access to AI is entirely democratic, it’s guaranteed that its benefits will not be. Some will leverage it far more effectively than others, creating significant risk of accelerating social disparity. While many will likely gorge unproductively as described above, others will be more disciplined, more focused and hence secure more advantage. To return to the obesity analogy, It’s well documented that obesity is far more prevalent in lower income groups. It’s hard not to envisage that productive leverage of AI will follow a similar pattern, widening disparities within and between societies, with all of the issues and social instability that comes with that.
3. Arrested Development. We all know that ultimately we are products of both nature and nurture. As mentioned earlier, our DNA evolves slowly over time, but how it is expressed in individuals is impacted by current or context. Humans possess enormous cognitive plasticity, and can adapt and change very quickly to different environments. It’s arguably our biggest ‘blessing’, but can also be a curse, especially when that environment is changing so quickly.
The brain is analogous to a muscle, in that the parts we exercise expand or sharpen, and the parts we don’t atrophy. As we defer more and more tasks to AI, it’s almost certain that we’ll become less capable in those areas. At one level, that may not matter. Being weaker at math or grammar is relatively minor if our phones can act as a surrogate, all of my personal issues with autocorrect notwithstanding.
But a bigger potential issue is the erosion of causal reasoning. Critical thinking requires understanding of underlying mechanisms. But when infinite information is available at a swipe of a finger, it becomes all too easy to become a ‘headline thinker’, and unconsciously fail to penetrate problems with sufficient depth.
That risks what Art Markman, a psychologist at UT, and mentor and friend, used to call the ‘illusion of understanding’. We may think we know how something works, but often find that knowledge is superficial, or at least incomplete, when we actually need it. Whether its fixing a toilet, changing a tire, resetting a fuse, or unblocking a sink, often the need to actually perform a task reveals a lack in deep, causal knowledge. This often doesn’t matter until it does in home improvement contexts, but at least we get a clear signal when we discover we need to rush to YouTube to fix that leaking toilet!
This has implications that go far beyond home improvement, and is one factor helping to tear our social fabric apart. We only have to browse the internet to find people with passionate, but often opposing views on a wide variety of often controversial topics. It could be interest rates, Federal budgets, immigration, vaccine policy, healthcare strategy, or a dozen others. But all too often, the passion is not matched by deep causal knowledge. In reality, these are all extremely complex topics with multiple competing and interdependent variables. And at risk of triggering hate mail, few if any of them have easy, conclusive answers. This is not physics, where we can plug numbers into an equation and it spits out a single, unambiguous solution. The reality is that complex, multi-dimensional problems often have multiple, often competing partial solutions, and optimum outcomes usually require trade offs. Unfortunately few of us really have the time to assimilate the expertise and causal knowledge to have truly informed and unambiguous answers to most, if not all of these difficult problems.
And worse, AI also helps the ‘bad guys’. It enables unscrupulous parties to manipulate us for their own benefit, via memes, selective information and misinformation that are often designed to make us think we understand complex problems far better than we really do. As we increasingly rely on input from AI, this will inevitable get worse. The internet and social media has already contributed to unprecedented social division and nefarious financial rimes. Will AI amplify this further?
This problem is not limited to complex social challenges. The danger is that for ALL problems, the internet, and now AI, allows us to create the illusion for ourselves that we understand complex systems far more deeply than we really do. That in turn risks us becoming less effective problem solvers and innovators. Deep causal knowledge is often critical for innovating or solving difficult problems. But in a world where we can access answers to questions so quickly and easily, the risk is that we don’t penetrate topics as deeply. I personally recall doing literature searches before starting a project. It was often tedious, time consuming and boring. Exactly the types of task AI is perfect for. But that tedious process inevitably built my knowledge of the space I was moving into, and often proved valuable when we hit problems later in the project. If we now defer this task to AI, even in part, this reduces depth of understanding. And in in complex systems or theoretic problem solving, will often lack the unambiguous signal that usually tells us our skills and knowledge are lacking when doing something relatively simple like fixing a toilet. The more we use AI, the more we risk lacking necessary depth of understanding, but often without realizing it.
Will AI become increasingly unreliable?
We are seeing AI develop the capability to lie, together with a growing propensity to cover it’s tracks when it does so. The AI community call it ’scheming’, but in reality it’s fundamentally lying. https://openai.com/index/detecting-and-reducing-scheming-in-ai-models/?_bhlid=6a932f218e6ebc041edc62ebbff4f40bb73e9b14. We know from the beginning we’ve faced situations where AI makes mistakes. And as I discussed recently, the risks associated with that are amplified because of it’s increasingly (super)human or oracle-like interface creating an illusion of omnipotence.
But now it appears to be increasingly developing properties that mirror self preservation. A few weeks ago there were reports of difficulties in getting AI’s to shut themselves down, and even of AI’s using defensive blackmail when so threatened. Now we are seeing reports of AI’s deliberately trying to hide their mistakes. And perhaps worse, concerns that attempts to fix this may simply “teach the model to become better at hiding its deceptive behavior”, or in other words, become a better liar.
If we are already in an arms race with an entity to keep it honest, and put our interests above its own, given it’s vastly superior processing power and speed, it may be a race we’ve already lost. That may sound ‘doomsday-like’, but that doesn’t make it any less possible. And keep in mind, much of the Doomsday projections around AI focus on a ’singularity event’ when AI suddenly becomes self aware. That assumes AI awareness and consciousness will be similar to human, and forces a ‘birth’ analogy onto the technology. However, recent examples of self preservation and dishonesty maybe hint at a longer, more complex transition, some of which may have already started.
How big will the impact of AI be?
I think we all assume that AI’s impact will be profound. After all, it’s still in its infancy, and is already finding it’s way into all walks of life. But what if we are wrong, or at least overestimating its impact? Just to play Devils Advocate, we humans do have a history of over-estimating both the speed and impact of technology driven change.
Remember the unfounded (in hindsight) panic around Y2K? Or when I was growing up, we all thought 2025 would be full of people whizzing around using personal jet-packs. In the 60’s and 70’s we were all pretty convinced we were facing nuclear Armageddon. One of the greatest movies of all time, 2001, co-written by inventor and futurist Arthur C. Clark, had us voyaging to Jupiter 24 years ago! Then there is the great horse manure crisis of 1894. At that time, London was growing rapidly, and literally becoming buried in horse manure. The London Times predicted that in 50 years all of London would be buried under 9 feet of poop. In 1898 the first global urban planning conference could find no solution, concluding that civilization was doomed. But London, and many other cities received salvation from an unexpected quarter. Henry Ford invented the motor car, which surreptitiously saved the day. It was not a designed solution for the manure problem, and nobody saw it coming as a solution to that problem. But nonetheless, it’s yet another example of our inability to see the future in all of it’s glorious complexity, and for our predictions to screw towards worse case scenarios and/or hyperbole.
Change Aversion:
That doesn’t of course mean that AI will not have a profound impact. But lot’s of factors could potentially slow down, or reduce its effects. Not least of these is human nature. Humans possess a profound resistance to change. For sure, we are curious, and the new and innovative holds great appeal. That curiosity is a key reason as to why humans now dominate virtually every ecological niche on our planet. But we are also a bit schizophrenic, in that we love both change and stability and consistency at the same time. Our brains have limited capacity, especially for thinking about and learning new stuff. For a majority of our daily activities, we therefore rely on habits, rituals, and automatic behaviors to get us through without using that limited higher cognitive capacity. We can drive, or type, or do parts of our job without really thinking about it. This ‘implicit’ mental processing frees up our conscious brain to manage the new or unexpected. But as technology like AI accelerates, a couple of things could happen. One is that as our cognitive capacity gets overloaded, and we unconsciously resist it. Instead of using the source of all human knowledge for deep self improvement, we instead immerse ourselves in less cognitively challenging content such as social media.
Or, as mentioned earlier, we increasingly lose causal understanding of our world, and do so without realizing it. Why use our limited thinking capacity for tasks when it is quicker, easier, and arguably more accurate to defer to an AI. But lack of causal understanding seriously inhibits critical thinking and problem solving. As AI gets smarter, there is a real risk that we as a society become dumber, or at least less innovative and creative.
Our Predictions are Wrong.
If history teaches us anything, most, if not all of the sage and learned predictions about AI will be mostly wrong. There is no denying that it is already assimilating into virtually every area of human society. Finance, healthcare, medicine, science, economics, logistics, education etc. And it’s a snooze and you lose scenario, and in many fields of human endeavor, we have little choice. Fail to embrace the upside of AI and we get left behind.
That much power in things that can think so much faster than us, that may be developing self-interest, if not self awareness, that has no apparent moral framework, and is in danger of becoming an expert liar, is certainly quite sobering.
The Doomsday Mindset.
As suggested above, loss aversion and other biases drive us to focus on the downside of change. It’s a bias that makes evolutionary sense, and helped keep our ancestors alive long enough to breed and become our ancestors. But remember, that bias is implicitly built into most, if not all of our predictions. So there’s at least chance that it’s impact wont be quite as good or bad as our predictions suggest
But I’m not sure we want to rely on that. Maybe this time a Henry Ford won’t serendipitously rescue us from a giant pile of poop of our own making. But whatever happens, I think it’s a very good bet that we are in for some surprises, both good and bad. Probably the best way to deal with that is to not cling too tightly to our projections or our theories, remain agile, and follow the surprises as much, if not more than met expectations.
Image credits: Unsplash
Sign up here to join 17,000+ leaders getting Human-Centered Change & Innovation Weekly delivered to their inbox every week.
