Artificial intelligence can be defined in a few ways, but I when I say AI in my article, I want it to refer to computers that can mimic human behavior well. Artificial intelligence is something different that consciousness, because “the assumption that if two systems are functionally indistinguishable, they will be mentally indistinguishable” is certainly false. AlphaGo is certainly functionally indistinguishable from a human playing Go, but it isn’t mentally indistinguishable. AlphaGo is a clever combination of Monte Carlo Tree search and constitutional neural nets to score boards. I think it would be difficult to argue that a computer that do these two algorithms/modeling techniques is considered conscious. But you can consider them functionally equivalent, so I would say that AlphaGo is still Artificial Intelligence in the broader sense of the word. I think right now we don’t have a very good sense of what qualifies as consciousness either. “We Can Quantify Machine Conciousness” puts forth integrated information theory as a way to say whether something is conscious or not, but I think this argument has a lot of holes in it. My main problem is that ITT is based on axioms that are supposed to capture conscious experience, but there is no real validation of whether these axioms capture what consciousness is. I think that is a very tricky thing that needs to rest on some lower level axioms, not just saying that these axioms are in themselves self evident.
I think perhaps a more practically relevant discussion for our current technological state is what role we should give artificial intelligence in our lives. We’re used to deterministic, well understood computer systems making decisions and being an integral part of our life. But artificial intelligence is inherently statistical/probabilistic in nature, so it is more difficult for us to accept something that works well a certain percentage of the time. This poses a lot of new risks for us. I think the Future of Life Institute’s letter for research priorities for robust and beneficial artificial intelligence is a good thing. Artificial intelligence is a new, powerful statistics tool that we need to consider how to manage effectively. The letter highlights the potential benefits that such a power can bring, but it is also important to consider the misuse of this technology. I’m not particularly worried about the end of humanity from AI, but it could have negative side effects we don’t intend. AI trained on tainted or biased data sets, behavior that optimizes a loss function but is unacceptable moral behavior, etc. are all concerns that must be researched and addressed. I think we need to be careful about where we integrate AI into our lives until technology has reached the appropriate threshold. For now, it is fine to let AI book dinner. But letting AI make court rulings is not an appropriate place to let AI into our lives.
There is also some concern about the economic effect of integrating AI. I don’t think anyone can really predict what is going to happen well. The Freakonomics podcast had two economists with very different viewpoints on the effect of automation and I think only time will tell which of those arguments is correct. I tend to put a higher probability on the market weathering the shock. People have an innate drive to do something worthwhile with their lives and I suspect we will have entirely new classes of things to do when AI is more prevalent in our society. It will take a lot of roles humans currently occupy, but there is always a bunch of things we never dreamed we wanted just waiting to be done. I also think there is an underestimation of the value that AI working with humans can provide. If we realize this value and do a good job of implementing AI that works for and with people instead of just replacing them, I think we’ll see even better results than if we just tried to replace people. As more experimentation is done with integrating AI into the workforce, I think this will be discovered. We will shift our thinking paradigm from AI replacing people to AI enhancing people.