Robot souls and the junk code of life

Dr Eve Poole on the concept of a robot soul and the importance of reflecting human agency, values, and ‘junk code’ in developing advanced artificial intelligence.

Two of the most pressing challenges in AI are the Alignment Problem and the Control Problem. These core problems are presaged in the plots of every Sci-Fi film you’ve ever watched: how to build robots that will remain aligned with human values, and what to do if they go rogue. But one route to solution that is in danger of being overlooked is to examine how these problems have already been addressed in the design of humans.

When you take a close look at our own blueprint, you notice a suite of capacities that look so clumsy we have generally been at pains to keep them well away from AI. The most famous is the messiness of our emotions, followed by our unshakeable ability to keep on making mistakes. We are also inclined to tell stories, are attuned to an uncanny sixth sense, and have an amazing capacity to cope with uncertainty. We have a persistent sense of our own agency in the shape of free will, and a propensity to see meaning in the world around us. I call these attributes features of the human soul.

When you consider why we have all these flaky and whimsical properties, you can glimpse beneath them a coordinated attempt to keep our species safe. For any species to survive, evolution demands that it has learning and adaptation as a superpower. In human design, this seems particularly facilitated by this set of capacities. Since the Enlightenment we have tended to write them off as a rather embarrassing set of junk code. But looking afresh, we can trace something in them that resembles a range of ameliorators with a common theme: they all locate individual humans within a community in order that there might be safety in numbers.

For instance, our emotions are there to help us relate better in relationships, because emotions are vital to promote healthy community living. Our accident-prone nature leads us to develop a conscience to avoid making harmful mistakes. This propensity for error makes us seek out others for guidance, counsel and comfort. This homing instinct for community acts as a regulating mechanism, to limit damage and to share learning for the benefit of the whole. Humankind consolidates this through storytelling, which cements identity and transmits life lessons, again within community. Our intuition or sixth sense allows us to access knowledge that otherwise seems to dwell in the realm of the collective unconscious, drawing on the wisdom of communities both of the present and the past. Uncertainty is so uncomfortable that we gravitate towards leaders to help us navigate through the fog, which again forms us into communities.

All of this ‘junk’ code promotes co-operation, and the kind of reciprocal altruism that creates the necessary conditions for our survival.

My concern is that in trying so hard to keep all this suspect code away from AI, we have inadvertently programmed a super-race of psychopaths, because this is the definition of any creature without a conscience. If we were to rediscover and celebrate our own junk code, we might be able to share it with those machines we have made in our image, but this time with the checks and balances from which we ourselves benefit.

It is a route fraught with complexity and danger, and one that requires an early conversation on robot rights. It is the route I think we need to take, not only to try to solve the Alignment and Control problems, but because it is the most human thing to do.