On Value Drift

The outcomes within any space-time region can be seen as resulting from 1) preferences of various actors able to influence the universe in that region, 2) absolute and relative power and influence of those actors, and 3) constraints imposed by the universe. Changes in outcomes across regions result from changes in these factors.

While you might mostly approve of changes resulting from changing constraints, you might worry more about changes due to changing values and influence. That is, you likely prefer to see more influence by values closer to yours. Unfortunately, the consistent historical trend has been for values to drift over time, increasing the distance between random future and current values. As this trend looks like a random walk, we see no obvious limit to how far values can drift. So if the value you place on the values of others falls rapidly enough with the distance between values, you should expect long term future values to be very wrong.

What influences value change?
Inertia – The more existing values are tied to important entrenched systems, the less they change.
Growth – On average, over time civilization collects more total influence over most everything.
Competition – If some values consistently win key competitive contests, those values become more common.
Influence Drift – Many processes that change the world produce random drift in agent influence.
Internal Drift – Some creatures, e.g., humans, have values that drift internally in complex ways.
Culture Drift – Some creatures, e.g., humans, have values that change together in complex ways.
Context – Many of the above processes depend on other factors, such as technology, wealth, a stable sun, etc.

For many of the above processes, rates of change are roughly proportional to overall social rates of change. As these rates of change have been increased over time, we should expect faster future change. Thus you should expect values to drift faster in the future than then did in the past, leading faster to wrong values. Also, people are living longer now than they did in the past. So even past people didn’t live long enough to see big enough changes to greatly bother them, future people may live to see much more change.

Most increases in the rates of change have been concentrated in a few sudden large jumps (associated with the culture, farmer, and industry transitions). As a result, you should expect that rates of change may soon increase greatly. Value drift may continue at past rates until it suddenly goes much faster.

Perhaps you discount the future rapidly, or perhaps the value you place on other values falls slowly with value distance. In these cases value drift may not disturb you much. Otherwise, the situation described above may seem pretty dire. Even if previous generations had to accept the near inevitability of value drift, you might not accept it now. You may be willing to reach for difficult and dangerous changes that could remake the whole situation. Such as perhaps a world government. Personally I see that move as too hard and dangerous for now, but I could understand if you disagree.

The people today who seem most concerned about value drift also seem to be especially concerned about humans or ems being replaced by other forms of artificial intelligence. Many such people are also concerned about a “foom” scenario of a large and sudden influence drift: one initially small computer system suddenly becomes able to grow far faster than the rest of the world put together, allowing it to quickly take over the world.

To me, foom seems unlikely: it posits an innovation that is extremely lumpy compared to historical experience, and in addition posits an unusually high difficulty of copying or complementing this innovation. Historically, innovation value has been distributed with a long thin tail: most realized value comes from many small innovations, but we sometimes see lumpier innovations. (Alpha Zero seems only weak evidence on the distribution of AI lumpiness.) The past history of growth rates increases suggests that within a few centuries we may see something, perhaps a very lumpy innovation, that causes a growth rate jump comparable in size to the largest jumps we’ve ever seen, such as at the origins of life, culture, farming, and industry. However, as over history the ease of copying and complementing such innovations has been increasing, it seems unlikely that copying and complementing will suddenly get much harder.

While foom seems unlikely, it does seems likely that within a few centuries we will develop machines that can outcompete biological humans for most all jobs. (Such machines might also outcompete ems for jobs, though that outcome is much less clear.) The ability to make such machines seems by itself sufficient to cause a growth rate increase comparable to the other largest historical jumps. Thus the next big jump in growth rates need not be associated with a very lumpy innovation. And in the most natural such scenarios, copying and complementing remain relatively easy.

However, while I expect machines that outcompete humans for jobs, I don’t see how that greatly increases the problem of value drift. Human cultural plasticity already ensures that humans are capable of expressing a very wide range of values. I see no obviously limits there. Genetic engineering will allow more changes to humans. Ems inherit human plasticity, and may add even more via direct brain modifications.

In principle, non-em-based artificial intelligence is capable of expressing the entire space of possible values. But in practice, in the shorter run, such AIs will take on social roles near humans, and roles that humans once occupied. This should force AIs to express pretty human-like values. As Steven Pinker says:

Artificial intelligence is like any other technology. It is developed incrementally, designed to satisfy multiple conditions, tested before it is implemented, and constantly tweaked for efficacy and safety.

If Pinker is right, the main AI risk mediated by AI values comes from AI value drift that happens after humans (or ems) no longer exercise such detailed frequent oversight.

It may be possible to create competitive AIs with protected values, i.e., so that parts where values are coded are small, modular, redundantly stored, and insulated from changes to the rest of the system. If so, such AIs may suffer much less from internal drift and cultural drift. Even so, the values of AIs with protected values should still drift due to influence drift and competition.

Thus I don’t see why people concerned with value drift should be especially focused on AI. Yes, AI may accompany faster change, and faster change can make value drift worse for people with intermediate discount rates. (Though it seems to me that altruistic discount rates should scale with actual rates of change, not with arbitrary external clocks.)

Yes, AI offers more prospects for protected values, and perhaps also for creating a world/universe government capable of preventing influence drift and competition. But in these cases if you are concerned about value drift, your real concerns are about rates of change and world government, not AI per se. Even the foom scenario just temporarily increases the rate of influence drift.

Your real problem is that you want long term stability in a universe that more naturally changes. Someday we may be able to coordinate to overrule the universe on this. But I doubt we are close enough to even consider that today. To quote a famous prayer:

God, grant me the serenity to accept the things I cannot change,
Courage to change the things I can,
And wisdom to know the difference.

For now value drift seems one of those possibly lamentable facts of life that we cannot change.

GD Star Rating
loading...
Tagged as: , ,
Trackback URL: