The Existential Threat of AI Consciousness | IPWatchdog Unleashed

This week on IPWatchdog Unleashed we explore whether artificial intelligence (AI) technology has progressed to the point where it has already achieved consciousness. In a nutshell, our panel of technologists do not believe AI is very close to achieving consciousness, but that it is indeed possible for AI to reach the point of consciousness, and to even reach the point of self-reflection, which would pose an existential threat to humanity.

Our conversation this week is from a panel presentation titled “Artificial Intelligence Today: A Discussion of the Technical Landscape of AI.” I moderated this conversation, which was between Jason Alan Snyder, who is Chief AI Officer for Momentum Worldwide, Malek Ben Salem, an AI expert, technologist and consultant, Dustin Raney, who is Head of Industry Strategy for Acxiom, and Dina Blikshteyn, who is partner and co-chair of the AI Practice Group at Haynes Boone.

If you want to jump finish and begin with the conclusion that AI consciousness poses an existential threat to humanity, you can watch this IPWatchdog Short where Snyder and Ben Salem discuss the particulars.

For those who like to start at the beginning, our conservation started with me asking whether AI has become sentient, and if not when we can expect AI will become sentient. This is a question I’ve asked Snyder each of the previous two years we have hosted an AI specific conference at IPWatchdog Studios. Two years ago, he predicted AI would become sentient within 15 years. Last year he predicted AI would become sentient within 14 years. Predictably perhaps, he agreed with his previous predictions and this year said, “13 years is probably a good guess,” but that AI is currently nowhere near achieving a sentient state today.  AI will become sentient at some point, but according to Snyder we are not anywhere near that currently.

“We’re not going to get to AGI, Artificial General Intelligence, through the systems, tools, and technologies that we have today,” Snyder said. “It’s going to require advancements… AI itself is a combinatorial innovation, and we’re going to require other technologies to help us get to that sentience. When we see innovations like we recently saw—Microsoft created a new state of matter for their quantum. When we see advances in computational biology and biological computing, those are the types of technologies which, although are not AI, are directly related to AI to help us get to that leapfrog event where we can get to a kind of consciousness with AI. So, the short answer is we can’t get there with the avenues that we’re pursuing today, and we’re still a long way away from where we need to be with those things.”

“Even though [AI] may seem like it has an understanding… it has limited memory and when it’s interacting with us it also interacts within a context window, so that memory is limited,” Malek Ben Salem pointed out. “So, I don’t see [AI] as conscious unless we get it to a point where it has enough memory that it basically develops experiences and accumulates experiences like we do as humans over 60 or 70 years, and we’re not at that point today.”

As the conversation unfolded, we spoke about whether hallucinations continue to be a problem for AI, whether the Turning test remains relevant with respect to defining AI, and fundamental aspects of what it means to be human. This brought us to a point where Snyder pointed out that AI “remix data, full stop, that’s all they do. Human beings interpret meaning… So, when these systems can interpret meaning, then it’s going to get really interesting.”

This provoked Dustin Raney—who in addition to being an AI expert is a guitarist—to discuss the creation of music by AI; particularly singing to compliment his guitar play. “It’s hysterical, what it’ll come out with sometimes. I mean, it’s somewhere around the ballpark, but it’s never what I would release out in the open.” This back and forth between Snyder and Raney led me to an epiphany, which I share during out conversation. AI is never going to satisfactorily product music—never.

“I don’t think we ever have anything to worry about, like ever,” I said as a lead into a story about some extremely talented professional musicians I had the opportunity to briefly work with earlier in my career. “There’s a difference between playing notes and playing music. I mean, a gigantic [difference], and you all probably have had that experience… you listen to something, it’s like, oh, that’s good, and then you listen to somebody who’s got real skill, and that’s different. And that seems to be what we’re talking about here with AI. They may be able to play the notes, but is it going to be able to feel? So, like the experience I just had retelling this story, I literally got chills when I think of that. And that’s what I think music and art should bring about—emotion. AI’s never going to have chills when they hear music or when they talk about it… It’s never going to be sentient; not the way that we feel sentient. Maybe it’ll be sentient like Data from Star Trek, right? Who realizes it has limitations but can’t go that extra mile to do any more than play the note.”

As we eventually wind into the wrap up we arrive at the point where Snyder and Malek Ben Salem discuss how AI could become an existential threat to humanity.

“If you unleash something into the world that portends to have some sort of consciousness, but without the ability to self-reflect, then you start to very quickly diverge into that sociopathic category,” Snyder explained. “And not a whole lot of great stuff is going to happen from that.”

“When it gets to that point [of self-reflection and self-consciousness], then it reaches a point where it wants to preserve itself, and potentially, through any means,” Salem explained. “And when it reaches that point, it may start shutting down other systems, AI autonomous systems, or doing other things that may harm, that may impact us as humans.”

“If we get [to the point of self-reflection and self-consciousness], we have to restrict its memory,” Snyder said in a matter-of-fact way. “It cannot have persistent memory if we get to the point that you’re describing. Otherwise, it becomes an existential threat.”

And on that happy note our podcast ends and the live panel transitioned into taking more questions from our studio audience. You can listen the entire podcast episode by downloading the podcast wherever you normally access podcasts, by visiting IPWatchdog Unleashed on Buzzsprout or by watching the conversation on the IPWatchdog YouTube channel.

More IPWatchdog Unleashed

For more IPWatchdog Unleashed, see below for our growing archive of previous episodes.