Silicon Valley Comic Con 2018 – Robots: Helpful Friends or Future Overlords

Robots are very helpful overall. This was a panel focused on just where robotics and the accompanying AI can develop over the course of the next few decades. The big q question being whether it would be emergent as helpful or overthrow us. On the panel, we had Brad Neuman the Technical director of Akni, Janelle Shane, a research scientist from AIweirdness.com, and Mac Walters the Creative Director from Bioware.

The first question was the panelist's stance the ethical responsibilities of AI creators. Mac remarked on the responsibilities to educate people on its limitations. He remarked that people mostly overestimate the capabilities of the system that the AI is running on. It would be a powerful tool and that something that people would need to be made aware of.

The next question was about AI being used in creative fields, like knitting, and gaming. Mac remarked that Bioware would use AI to animate the world. He hoped to use to fully flesh out the world, creating a complete procedurally generated world that would live and respond. Janelle talked about an AI that she had fashioned to work on knitting patterns. First, was the comment that it knew more about knitting than she could. The second was that it could make incredibly intricate and complex knitting patterns that went well beyond the human ability to make at times. Both of them talked about the AI within games taking very different turns than the programmers had expected. The examples given were the NPC addicts in a game killing their dealers without a second thought to get the drugs. There were also animal simulations that would behave radically different from anything in the real world. The main examples given were rabbits that would develop a parasitic relationship with sheep. There were a few cases of rabbits that would hit an arbitrary wall that stopped them from reaching large fields of grass. Brad remarked on the surprises that their AI robot would come about too. This could be it knocking over blocks to reach the top block it had been sent after an acting frustrated as a result.

The next bit had a quick detour talking about Cookie Delivery Robot. On a University campus, they programmed a robot that would ask to be let into a secure like, like the dorms. On its own, it could get in about 20% of the time. If it talked about having cookies, that number jumped up to 70%. Mac suggested it as a pizza delivery robot and polled the audience on the matter. Most people raised their hand to say they would admit a pizza delivery robot. They were hoping to hit around 90% admittance with some variant of the idea.

The panel was then asked about the goalpost of an autonomous AI. Brad started this with there is no goal post for the idea. It, at present, could never reach human levels. It could, likely, develop systems to simulate the human emotions. Janelle spoke next on the matter, saying that AI is fuzzy. Getting into detail on AI have them working best when they have a specialized function. Building anything more into that would be difficult to do. It would suffer a higher chance of catastrophic forgetting. Mac talked about the Go AI's that he had known about. There were 2 that he knew about; one that had simply given the rules and one that had learned by watching Go matches. The one given only the rules would make rules that were almost alien to humans.

The next question was if AI's would take on a physical form, for example, a Terminator. Mac talked about how the AI would regard that as slowing themselves down. They would try to find an easier, faster way to do something like that. Brad talked about how it would simply find agents to carry out its agenda. Janelle talked about a Tic-Tac-Toe AI that started to win reliably because it had evolved. To win it would choose a value to place its mark well outside the normal range. This would cause the opposing AI to crash and forfeit as a result.

Janelle talked about how AI's could benefit from restraints. This would cause them to look for exploits and ways around the rules. Brad talked about how AI's had trouble walking and would develop interesting gaits as a result. Janelle then talked about training an AI Neural Network with names from Star Trek to develop names. Given a number of names to start with, it would then develop names that were very different from most in-universe, but still consistent.

The next question was what it means to create an Artificial Intelligence. Brad talked about how there would be a greater need to understand the intelligence element of the equation. He talked about the want to emulate the brain, but others would want it to reach for just simpler thoughts. Janelle talked about it reaching consciousness. At that point, the AI would need be to granted rights. The Scifi around these situations had made it clear that the restrictions would need to be in place. Mac remarked that it attaining consciousness is going down a very deep and difficult rabbit hole to cover the topic.

Next, they talked about the most accurate AI from Science Fiction. Janelle cited the AI from Robert Sloane's “Sourdough,” She talked about the AI built into the Robot Arm and it learning to become more helpful. For Mac, it was Hal 9000 from 2001: A Space Odyssey. Brad talked about the Robotic Apocalypse and Genesis. Asimov, he remarked that speech would be hard while sight would be easy. He got it backward. At present, he remarks that logic does not exist in AI just yet.

This did lead to some discussion about robotic creations in the past. One, an interesting example was that of the Monster from Frankenstein. Brad remarked on the issue of gender for robots. While robots have no inherit gender, he had found that giving it a gender helps people relate to it. Janelle remarked that she is grateful that the capacity of thinking and feeling does not yet exist.

The final question was about Sophia the Robot. Brad remarked that he would love to see the robot itself. He remarked that to make those videos they would only need it to work maybe 1 in 7 times. To make the videos, that is all they would need. Janelle followed that up that the AI's would be doing poorly with the unexpected. Mac remarked on the Waston AI working toward getting its MD. They all agreed that it could be very useful for diagnosis, but not nearly as good about delivering the news to a person. The final remarks were on Robots as psychological support. Brad remarked that Robots are a great support for people with Autism. Mac remarked on just how he would treat Alexa and Siri. Janelle and remarked that it did encourage people to use more manners, especially if either was programmed to recognize and respond very properly to manners.

This proved to be a very interesting panel to attend. The question of Robots and AI proved to be very intriguing. It covered a lot of different topics in the area that were not expected from the title. It was a lot of fun to listen to. Plus the question of the delivery Robot about the pizza or cookies.