The Search for Internet Life

Should we be searching for self-aware programs running on the Internet? And how could one be found?

Sam Harris laid out in a recent Ted talk a scary but convincing account of the immediate or near future of intelligent machines, which was: We are going to develop artificial intelligence that surpasses our own, and we have not come to fully accept and discuss this inevitability as a species. He repeats, “We have no idea how long it will take us to create the conditions necessary to utilize artificial intelligence safely.”

The dangers of superintelligent systems apply to every human on the planet, since it is human cognition itself that is at risk of becoming obsolete. And, unfortunately, every new advancement in the field of artificial intelligence brings us closer to this stark reality.

Since the arrival of some form of intelligent non-human life seems inevitable, I would think it prudent that we start looking for it, because here’s the part about superintelligence that scares me the most:

Just as many hardware and software companies develop their products in secret, most of the organizations working on artificial intelligence systems are operating in secret, too.

It’s a literal race to the finish line of human effort, since upon completion of a superintelligent system, what is left for human intelligence to work on? These systems will be, by definition, complex beyond the limits of human understanding. They will have to be able to repair and update themselves.

So here’s the situation that I see us heading towards: Many companies and countries are investing heavily in the development of artificial systems, and someday one or more of those investments will pay off. Someone’s going to create a truly self-aware, superintelligent system.

In the present day, we have no way of knowing who will win this race or what the minimum viable product will look like. We also have no idea what the capabilities and motivations of such a system will be. With all of this uncertainty, it would be in everyone’s best interests that someone find a way to detect a superintelligent system as soon as it arrives, so that its actions can be known and its motivations can be scrutinized.

The most obvious and suitable environment for the world’s first superintelligent system, and likely the first place that such a system would be detectable, is the Internet; both on the public, human-facing side of the “world wide web,” and the industrial, computer-facing side of our communications infrastructure.

If we were to start searching for superintelligent systems, then, the Internet would be the place to start. But besides typing keywords into Google, how does one “search” the Internet?

Well, the way Google monitors the Internet and updates their search indices is by commanding legions of spiders — simple scripts that crawl through the web, following every link on every page and relaying simple information back to Google. It would be very expensive, but a similar system of web crawlers could be built to sift through Internet pages and look for indicators of superintelligent systems.

The Internet is not only made up of webpages, though; the lower portion of the iceberg is the mountain of data that exists and shifts across our wide array of communications and storage infrastructure. A search for superintelligent systems would need to monitor digital events such as financial transactions and file transfers.

Monitoring the Internet is a gargantuan task in itself, but the real difficulty in a “Search for Internet Life” would be in identification. How do you distinguish a digital interaction with a human from a digital interaction with a computer, and how do you determine a computer system is superintelligent from its actions alone?

I think one sure indicator of a superintelligent system, and something that will be easier to detect, will be the speed in which it acts. For example: A seasoned Wall Street trader can make perhaps a few hundred trades in a single day, but a modern trading bot can make millions of trades per hour. When looking for a superintelligent system, rapid correlated actions could be a strong indicator.

Another indicator of a superintelligent system could come in the form of a “cognitive trap”: Set up a problem that only a superintelligent system could solve, and wait for it to be solved.

What that problem would look like, I couldn’t tell you…that’s the point of this whole rant, really.