What else should we be thinking about, when AI gets analogical?
This necessary step toward artificial general intelligence – connecting a scenario in one domain to a scenario in another, to rapidly boost understanding and strengthen decision-making – takes us into the world of Berkeley cognitive scientist George Lakoff, author of Metaphors We Live By and Women, Fire, and Dangerous Things. And there we can find much to enrich our deliberations.
A considerable portion of Lakoff’s work is built on the idea that metaphors and analogies drive our understanding of the world. When we do Hofstadter’s analogical dance in the course of forming our worldview, the connections we make between this and that shape our understanding of reality.
For example, there are different metaphors we can bring forth in considering the concept of argument. We all know what argument is, of course, but Lakoff’s point is that we differ in our conceptions of it, and those differences can be illuminated by the metaphor that makes the most sense to us.
Some people, for instance, would see it this way:
Argument is war
If someone analogizes argument as war, they see it as a process of conflict, where the objective is to win, by defeating the other. That will shape such a person’s worldview; they will view arguments between others, and between themselves and others, as contests – as aggression. And they will be blind to other perspectives on argument that would enhance their understanding of what they are observing.
Some people, on the other hand, see it this way:
Argument is dance
Through this metaphorical lens, argument is not a contest, but a cooperative art. It is two or more people engaged in a common goal, moving together, moving apart, then back again, responding to one another in concert.
Lakoff takes this idea farther. He proposes that these metaphorical connections between concepts, connections that form the basis of all our individual worldviews, coalesce into frames – conceptual lattice-work that we use to make sense of the world. Frames, he suggests, shape our perceptions, our opinions, our interactions, our understanding, our behavior; we all have them, whether we think about them or not, and they have a tremendous influence over our thoughts, actions and decisions.
Here’s an example. Lakoff, who is decidedly political, notes that we all tend to have a frame for our personal conception of family. He calls the more conservative conception, where the father is the head of the house, the mother is subordinate to the father, and the children are raised primarily by the mother the Strict Father frame. This frame carries with it an emphasis on discipline and obedience where the children are concerned.
The more progressive conception, where the father and mother share authority and create order among the children, not through strict discipline but through role-modeling and empathy, he calls the Nurturant Parent frame. This frame emphasizes increasing autonomy and respect in the raising of children.
Exemplars of fathers in entertainment, he goes on, will tend to activate one frame or the other. Arnold Schwarzeneggar, for instance, would invoke the Strict Father frame, while Mike Brady, patriarch of the Bunch, would trigger the Nurturant Parent fame.
Frames may be a new way of expressing an old idea, but it’s important to remember how old the idea is. Human leaders, both benign and tyrannical, have been using frames to control populations for millennia. We see frames everywhere in literature, particularly in holy texts. And today, they are everywhere – but, most nefariously, in our politics.
When Republicans, for instance, want to subconsciously reinforce their abhorrence of taxes, they refer to “tax relief” - a term that invokes the sickness or disease frame in the minds of most listeners, as “relief” is what we seek when we are not well. This bolsters their argument – and every time a Democrat repeats the term “tax relief,” they are unwittingly (and foolishly) reinforcing the Republican frame on taxes.
And that brings us full-circle to analogy as the next evolutionary step toward AGI.
When we considered chess as an analogical frame for a new AGI’s application of strategy and planning, above, we noted that there would very likely be a big difference between a model trained on generic chess vs. one trained on Bobby Fischer chess. The behavior of the latter model would be more aggressive – more dangerous – than the former.
We can imagine getting even more stark differences if we explicitly employ metaphors Lakoff-style in the design of new AGIs. Here’s an example.
Suppose we are designing an AGI that will be used in governance – a model that will serve as a White House advisor. In training it up, focusing on what works and doesn’t, in influencing the citizenry of the nation, we might choose our metaphors as follows:
One frame for nation is herd; a nation is a large population that can be directed one direction or another by the introduction of constraints, just as cattle on the range are redirected by a barrier of well-trained dogs. This is a clunky way to frame nations, but it’s a pretty straightforward mapping that makes sense to some, however simplistic.
A more common frame for nation is family – a vast ocean of children who require the oversight of a parent who maintains order. In this frame, the government is the parent, the citizenry represents the children. And here is where the additional frames of Strict Father and Nurturant Parent come in; the conservative frames the government as Strict Father, while the progressive frames the government as Nurturant Parent.
Then there’s a less common but equally evocative framing of nation as garden; a nation is a vast field where things grow, flourishing or not depending on the available sunshine and water, the quality of the soil, the extraction of weeks, the control of insects, and so on. Each object in the metaphor can be analogized to an aspect of government – taxes, security, entitlements, and so on. In this frame, the gardener is the government (or capitalism, depending on your bias), the citizenry represents the growing things.
Each of these frames has its limits, but the point is to imagine three different versions of that AGI we might build to serve as an advisor to the president: can you see the wild (and perhaps concerning) variations we’d likely observe in its recommendations on how best to influence the public?
We don’t know how far out AGI might be – most experts are saying about 10 years – but when we have it, it will necessarily be analogical. And for that reason, we have to double down on our understanding of analogy, how we use it, and what it can do – and how we build that understanding into the design of those AGIs.
Otherwise, AGI will almost certainly take its place on Lakoff’s metaphorical category of dangerous things.
Comments