Ken's Blog

When we look back in the fullness of time
There are certain moments we mainly recall

We don’t get to have many moments of grace
But they are what let us make sense of it all

Continuing the general theme from yesterday…

Suppose, sometime in the near future, you are I are working on a project together, and you become unavailable for a while. Perhaps we are collaborating from opposite sides of the planet, and you need to catch some sleep.

Suppose further that there is an A.I. that has been trained on you, so it does a fairly decent job of embodying your expert knowledge. It doesn’t have your ability to provide novel creative insights, but when asked questions about your domain of expertise, it does a good job of answering questions in your voice.

Should I continue working with your A.I., until you return? Would that ersatz experience seem normal / odd / unpleasant?

Or is this something that we will all soon take for granted? Will children grow up thinking that it has always been this way, and find it hard to imagine life without it?

I had a recent conversation at a conference about using A.I. agents for collaboration. We discussed many ideas, including what happens when one collaborator uses an A.I. proxy of the other collaborator. For example, if an invention comes out of that work, whose name(s) should be on the patent?

We also discussed two people working together while each has a team of A.I. agents at their fingertips, trained on expert knowledge that is seamless fed to the collaborator. How possible is that? And if it is feasible, would it give people some kind of collaborative superpower?

At one point one of the people in the conversation got really excited. He said “This would make a great research paper at the next conference. We could use an A.I. to help us write it. I think it would win the Best Paper award!”

“Yes,” I said “but I would be really sad if they give the award to the A.I. instead of to us.”

At a recent conference I saw a demo of an AI video-chat assistant, which appeared to participants as an on-screen cartoon character. The AI assistant would listen to the group discussion, and when it determined that the conversation was floundering, would jump in and try to refocus everyone, pointing out possible next directions based on what had already been said.

Unsurprisingly, the AI was articulate, organized, and well focused. It also had a good grasp of the topic and of possible next topics.

But I had a big problem with the very premise of the project.

In terms of clarity of expression, organized thinking, and ability to summarize ideas, an AI will usually do better than a human. After all, an AI is essentially an amalgam derived by training on prior examples of many humans who have shown how to exercise just those skills.

But an AI is not actually a human. It’s entire “world view” is based on what it has already seen. So if an idea appears which is completely novel, the AI likely won’t realize that it is seeing something new.

In short, a problem with having an AI moderator for an intellectual conversation between humans is that the AI, while generally being much more organized and articulate than the humans, may not recognize when the humans are actually stumbling toward a completely new idea.

I am going to be giving a talk at a workshop in a few hours. Right now I am putting together the final version of the talk, a combination of words, images and demos.

It occurs to me that the talk itself and the preparation for the talk are actually two separate events. Sure, everyone sees the talk that is given — that’s the official public event.

But the act of preparation and authorship is in a way a more subtle and interesting event. It involves not only the things that I end up saying, but also the things I end up not saying.

The act of pruning and editing defines the scope of the talk, and therefore the particular shape of its boundaries. What people do not see and hear, because it is merely interesting yet not essential, is a fundamental part the message.

As I an giving people a Soma cube to play with, I might as well give them something to do with it. Today’s post is a link to a more advanced version of my Soma cube visualizer.

In this version you can move the pieces around independently. If you are clever, you can create all sorts of cool 3D shapes.

Enjoy!

Whenever I travel halfway around the world, I find it hard to understand why some people continue to insist that the earth is flat.

On the other hand, I don’t understand why some people continue to think that our current US president is doing a good job.

Maybe there is a connection.

At some point communication and display technology will evolve to be able to support the widespread use of what might be called “Holographic Zoom”. Pretty much anybody will be able to have a remote face-to-face conversation that will feel as though they are sitting across a table from each other.

There are already early technologies that do something like this, but they require high end equipment. For example, Google’s “Beam” technology does it with 6 cameras and a special fancy 3D display.

But that’s not something available to millions of people. The real change would happen when such a capability becomes inexpensively available to everyone.

When that happens, I wonder how it will change things. Will people travel on airplanes less, because they will be able to have the experience of meeting face to face without the hassle of actual travel?

Or will people travel more, because they will feel more connected with one another? I guess only time will tell.

Suppose you could record everything that happens in the course of every day. Further suppose that you could have that information automatically indexed and played back at will, in a way that is contextually searchable (eg: discussions about movies).

On the one hand, this seems like a formidable super power. On the other hand, having the ability to forget about things — and the ability to truthfully say that you’ve forgotten about things — seems like something we would not want to lose.

Which would you choose?

I don’t want my students using AI to do their homework. That would defeat much of the purpose of learning to do things for yourself.

But I am starting to think that it would be beneficial for students to use AI to help them evaluate their work. Given that AI is (at least in its current form) essentially a patchwork gathering of the expertise of many humans, students could benefit from what those many human experts think of their work.

The trick will be to figure out how to invite students to use that resource without implicitly inviting them to use AI in the work itself. I wonder whether it would be possible to design homework assignments in a way that encourages the former while discouraging the latter.