When discussing how individuals and societies will be affected by AI, it is important to balance the benefits with the potential risks. The short-term benefits are ample and easily understood – AIs can take over dreary, repetitive jobs and free people to realize their potential, while getting algorithms involved in decision-making can limit both the errors and the biases that humans are prone to introducing. By leaving decisions to an algorithm, we can make sure that the innate human limitations – biases, insufficient information, moods – aren’t affecting decisions in an overt fashion. By introducing algorithmic logic, we can make sure that human frailty isn’t driving the big decisions that society needs to take.

That said, we often forget that AI has both short and long-term impacts on society. Looking at things in the short term, it may well look like AI is nothing but a net positive for society. AI can help us sort out issues such as suboptimal urban planning, or deal with racial bias in sentencing decisions. It can help clarify the impact of credit scores, or ensure that the mood of a doctor doesn’t affect a medical diagnosis. What unites these cases is that it is very easy to spot bias or errors in the way the AI functions. An AI that does urban planning in a way that marginalizes certain ethnic groups will be found out and an AI that misdiagnoses cancer will be caught. These are all cases of what I have called ‘short bias’, errors that algorithmic logic can get caught in through insufficient data or bad training.

But what about those cases where an AI influences decisions that have long trajectories, and where the impact might not be known for years or decades? Imagine that an AI is programmed to figure out which of four new research paths in energy production should be supported and financed. One is known and tested, two are cutting edge but with great potential, and the last one is highly speculative. Unless the AI has been programmed to take great risks, it is likely to suggest that the speculative program is cut. Yet, we know that many speculative ideas – antibiotics, the internet, and female suffrage come to mind – have turned out to be some of the best ideas we’ve ever had.

What is at play here is something I have given the name ‘long bias’, i.e. the issue of potential long-term negative consequences from AI decisions that are difficult to discern in the here and now. AI is exceptionally good at handling issues where the parameters are known – whether a cat is a cat, or whether a tumor is a tumor. These are also issues where humans can quickly spot the errors of an AI. When it comes to more complex phenomena, such as ‘innovation’ or ‘progress’, the limitations of algorithmic logic can become quite consequential. Making the wrong bet on a speculative technology (and let’s be clear, there was a time when the car was just that) can affect society not just in the here and now, but for a very long time afterwards. Cutting off an innovation trajectory before it has had a chance to develop is not merely to say no in the here and now; it is to kill every innovation that might have been, and an AI would not care.

In this sense, AI is a double-edged sword. It can be used to make decisions at a speed that no human can match, with more information than any group of humans could process. This is all well and good. On the other hand, by taking away the capabilities of imagination and bravery that humans excel at, we may be salting the earth for technologies we’ve not even considered yet. AIs work with data, and all data is historical – as the investment banks say, “past performance is no guarantee of future results”.
With this in mind, it is far too early to be wishing for a technological singularity, a state of affairs where infinitely wise AIs can guide us in our technology exploration. On the contrary, when it comes to innovation it is of critical importance that we retain the human capacity to imagine and dream, and ensure that we are not letting data do all the driving. AI can help us solve massively complex problems, but the keyword here is ‘help’. The human capacity “to see a world in a grain of sand/ and a heaven in a wild flower” needs to be protected, to ensure that AI only augments our capacity to innovate, rather than defining the same.