GroshLink

AI: A View from the Inside (Mike Bowling, 10/01)

Note: written for the Graduate Christian Fellowship's (GCF) vocation project (VP).  Now that the GCF site has been taken down, over time I'll post the pieces for your blessing.  Thank-you to all the alum who participated in this work.  Alumni friends, I would love to have to hear your current thoughts on your vocation.  Maybe we can have that conversation at the alumni reunion on April 21.  We'll have a 9am breakfast at the Seigfrieds, a 1pm reception most probably at the Adamson Wing, and a 6pm dinner at the Moores.  Email me for more details if you haven't received an invitation by email.  Current GCF activities can be found at http://www.u-connectpgh.org

Preliminaries¹

I'm not a philosopher. I'm not a historian. I have no intent on adding to the extensive discussion on the philosophic underpinnings or historical development of artificial intelligence (AI). There's enough good stuff that I need not say anything, and enough bad stuff that I dare not. What I do hope to contribute is a view from the inside. I am a Christian. I am also a doctoral student in the field of AI. I've been doing research for eight years, in the areas of computer vision, machine learning, robotics, and game theory. I also have the scandalous confession that I'm an AI pessimist as I believe many closet researchers are. Although I seek to push the cutting edge, I often consider the entire enterprise somewhat shaky.

Practical Views on Intelligence

The generally agreed upon, but rarely stated, goal of AI is building intelligent entities. But this goal is rarely considered in the daily routine. AI is a field that straddles the engineering and scientific disciplines. The goal of AI is to build something, and most researchers are currently in the act of building something. Yet the intermediate goals are scientific. We build things to understand general principles in order to build the real thing.  This straddling means that to do research in this field you rarely have to consider the ultimate goal. In fact, there's a good chance you could take an undergraduate AI class and spend less than five minutes talking about the real thing. Many researchers, including myself, spend little time thinking about it and even less time describing it. It's simply not practical nor original (as philosophers have already overzealously claimed the area as their own.)

At the same time we still hold implicit and operational beliefs about the definition of (artificial) intelligence. It's dangerous to try to categorize beliefs, but the more prominent views are that of the Intuit, Rationalist, Behaviorist, and Humanist. The intuit says, Intelligence is like pornography, you know it when you see it. The rationalist says, Intelligence is acting optimally, like playing the perfect game of Chess, investing to maximize return, etc. It has to do with humans only insofar as they tend to make rational decisions. The behaviorist says, Intelligence is what humans do. When a car drives itself it is exhibiting intelligence. The Turing Test is the real thing. Finally, the humanist says, It's what humans have. An airplane is not a bird, just because it flies.

The viewpoint of the humanist may need some additional clarification, as well as the related implication. A humanist would not consider the Chess playing program, Deep Blue, intelligent. It is no more than a very fancy and fast directed search. Although humans do search in Chess, grandmasters do incredibly little classical search, and they definately don't follow through moves to the depth that Deep Blue does. In essence, Deep Blue is cheating the intelligence problem. Since we have so little understanding of the nature of our own intelligence, and so great an understanding of machines, the implication is that machines cannot be intelligent without breakthroughs in the understanding of our own intelligence. This is supported by the fact that the true successes in AI often end in a different subfield of computer science (e.g., search is now the area of algorithms.)

The Connection To Humans

Any given researcher probably operates, and only subconsciously, under a mixture of these viewpoints. The intuit view gives a great deal of freedom, the rationalist can measure success, and the behaviorist and humanist seems to be at the real heart of the matter. AI is undeniably concerned with human intelligence. The two things are inseperable. Consider the following second paragraph of an AI textbook.

AI addresses one of the ultimate puzzles. How is it possible for a slow, tiny brain, whether biological or electronic, to perceive, understand, predict, and manipulate a world far larger and more complicated than itself? How do we go about making something with these properties? These are hard questions, but unlike the search for faster-than-light travel or an antigravity device, the researcher in AI has solid evidence that the quest is possible. All the researcher has to do is look in the mirror to see an example of an intelligent system. (Russell & Norvig, 1995)

This quote illustrates two things: AI is really about human intelligence, and the assumption required before you step on the field is naturalism, i.e., matter is all there is. We see naturalism in the assumption that humans and their brains are no more than a biological machine. But something odd is going on here. Proofs of our brains as just machines point to AI successes, and AI (as above) points to our brains as an existence proof of machine intelligence.² This creates a mind-numbing circular argument for those who think our brains may be more interesting than that.

The search for the right mechanism for intelligence seems to be what drives AI from one holy grail to the next. Consider the history of bandwagons that have advanced the field: symbolic AI and logic-based systems, knowledge-based systems, embodied AI and robotics, neural networks, learning, genetic algorithms, and probabilistic systems. All of these get at one aspect of human intelligence from knowledge to embodiment to neurobiology to evolution to learning. The community seems to be constantly believing that each breakthrough will be what's been missing. Maybe this is what caused the flamboyant ten-year predictions that were so common for the first few decades. It's almost like the face-in-the-mirror is not an encouragement that all this is possible, but a thorn that this should be easier. I should note that this is a slight exaggeration as many researchers are resigned to AI being a long haul.

A Christian Response

So what does all this mean for a Christian in the field? What does a Christian approach to AI look like? A Christian response to naturalism and mechanistic intelligence is only partially the issue. The bigger question is how should my faith affect my research? And I mean more than just academic and scientific ethics. Let me make a couple points toward finding a Christian approach to AI and then conclude with some unanswered questions.

I think one of the most crucial differences that a Christian brings to the field of AI is awe. We should regularly remind ourselves that our God is the awe-inspiring creator who did not make us like the horse or like the mule, which have no understanding. (Psalm 32:9; NKJV) The biggest success of AI may be its failures as we come to realize and appreciate God's creation in general and his special creation of man. Usually we see science demystifying something thought divine, but in AI there's almost a mystification as we struggle to understand how our minds do even the smallest of unconscious acts. I don't mean to say we should root our faith in our failings to build intelligence. I only mean that the face-in-the-mirror should fill us with awe and marvel at the work of our Creator. This awe should also be shared. The sheer difficulty of the intelligence problem is often depressing. Sometimes we struggle to add even the smallest incremental progress toward a goal that it seems should be so close. Awe, astonishment, praise, and worship, seems a refreshing alternative visible to our colleagues.

In more practical terms, I think Christians can skip the tendency to look for black-box solutions or biological plagiarism. I think the humanist view of AI has influenced solutions of a black box variety. Since we don't understand biological intelligence we look for mechnical systems with the same mysticism. For example, genetic algorithms is a good example of a technique that has received massive attention almost because of the lack of understanding in how and what is going on. Biological plagiarism is what I call the attempt to copy a biological entity into mechanical parts with the hope that intelligence will come with it. For example, neural networks in a loose way, much more in words than reality, attempt to do this. I think this may become more popular as our biological understanding and computational power increases. Basically, both of these approaches hinge on humans being an existence proof for intelligent machines.

My steps toward a Christian response really avoid the larger issues. Should we even bother with AI at all? Will AI go too far? Or can it inherently not go anywhere, so why waste our time? How else should the lack of a face-in-the-mirror proof affect how we build, develop, and search for intelligent mechanisms? Should nature be a valid source of inspiration and in what ways? Will robots ever be capable of enslaving humanity as power generators and will this lead to the endtimes?

---

¹ This was a presentation at the Graduate Chapter Focus Weekend track on integration of faith in the academy. So it is both unnecessarily formal and unnecessarily technical, for which I apologize. Or maybe this is just an excuse to get you to come ask me questions.
² It's even more common to hear justification for a particular technique appealing to our brains or nature. It seems particularly outrageous in the areas of neural networks and genetic programming.