Dr. Joscha Bach of the MIT Media Lab, and the Harvard Program for Evolutionary Dynamics, has dedicated much of his research pursuits to figuring out how the mind works. Over a decade ago he founded the MicroPsi project, in which virtual agents are constructed and used in a computer model to discover and describe the interactions of emotion, motivation, and cognition of situated agents. Bach’s mission to build a model of the mind is the bedrock research in the creation of Strong AI i.e. cognition on par with that of a human being.
Building Stronger AI with Reinforcement Learning
Reinforcement learning drives much of the agent interactions in MicroPsi. Though a type of machine learning, Bach points out that reinforcement learning is “different from machine learning, in that it involves interaction with the world and becoming more intelligent as a consequence, something that AI is not yet smart enough to do on its own”.
Humans and other intelligence beings look for external features that they can organize into hierarchies of information. The recognition that takes place in “the mind” of a computer is, in some part, based on a similar model. Scientists use a “grammatically-simplified” model, which allows the computer to recognize patterns, images, even actions and events. The agents are able to make sense by identifying suitable behavior, in turn making more sense of the world. They are able to organize the world just by seeing it and identifying statistical dimensions.
Humans do this constantly by taking signals from multiple dimensions – time, space, color, etc. – in real-time, forming neural networks that help us learn how, and how not to behave, in the world. We then organize these concepts into categories, such as features, object permanence, mental states of people, etc., and then come up with models and theories to explain these abstractions.
Joscha and other cognitive scientists believe that figuring out how the mind works will be the catalyst for success in reverse engineering a strong AI mind. “One of the big questions”, says Joscha, “is how much do we have to put into the machine in the first place?”
The Moneyed Hands that Pave the AI Road
On the road to stronger AI, Bach provides a no-frills perspective on how the direction of AI will be influenced in the decades ahead. Joscha thinks the strongest winds may not always come from the ‘makers’ in the lab and field, but from the ‘shakers’ at the top of the economic totem pole – the ones providing the funding. Right now, if you want to really make a quick and lasting change on the planet, you have a great chance of doing so in the form of an organization as opposed to an individual. “AI is likely to come from the top as an extension of business intelligence…it’s super dangerous if we make corporations more powerful, as they don’t always act in the best interest of humans”. In some ways, corporations already have more rights than human beings, and they operate much like agents do, in an efficient and goal-oriented manner, but often without a whole unified consciousness.
Bach’s prediction should be discussed publicly and privately amongst tech companies, private investors, and policy makers. Precedents in the grey areas of morality and progress have been set before, and history cannot be brushed away or ignored. For example, though the evidence against burning fossil fuels is rather clear today, companies are still harvesting and burning fossil fuels. Why? The commodity does well in the stock markets; if the companies stop supplying and using, the stock market and companies incur huge losses.
This approach by the corporate world ties into another looming – that AI will soon replace an unprecedented number of jobs. “If a person is no longer fulfilling your organization’s goal, and you (as an organization) want to survive, you’re going to replace that person by somebody else who is better aligned with the goals of the organization, there’s a kind of evolution going on among these organizations, and part of this evolution is how well they perform in the marketplaces, the financial systems, how well they are able to better other organizations to persist and stay…in power,” says Bach.
Companies in a wide range of industries are already starting to use AI, and at some point may not need humans. “…Companies don’t care about externalities, they care about the parts that make it more efficient and help it survive”, says Joscha. He believes the challenge of making a world safe against AI is embedded in our decisions to make economic decisions that are in line with the best human outcomes
Though AI poses risks, now almost synonymous with the Future of Life Institute open letter presented to the United Nation and signed by Musk, Gates, Hawking, and other leading thinkers, funding for AI will likely not stop anytime soon. Bach believes the only motivation for stopping funding would rest on the belief that a certain goal is not possible, not because we fear AI.
But if one day humans are no longer primary agents on the planet, and those “artificial” agents have goals that are not like ours, we might find ourselves in a hard spot. As Joscha explained, “Musk gave generously, for good purposes, to support AI research; at the same time, research into (both) development and risk are important. It makes technical sense for him to be concerned about the dangers as it relates to probabilities of our demise by AI”.