The Future of AI: From Quantum Computing to the Internet of Things

Sunday, 03 September 2017 - 7:15PM
Artificial Intelligence
Sunday, 03 September 2017 - 7:15PM
The Future of AI: From Quantum Computing to the Internet of Things
< >
Image credit: TriStar Pictures
One of the most well-attended and energetic panels at Escape Velocity this year was "Artificial Intelligence: How Close Are We...And How SAFE Are We?" The panel featured Nick Kelly, a cyberpunk author and cyber-security professional, Josh Baptist, an R&D specialist for eSkin, and Joe Haldeman, the bestselling sci-fi author of The Forever War (you can check out our story on his solo talk at EV 2017 here!).

Over the course of the hour, the discussion ranged from quantum computing and robotics to hacking and the ethics of creating sentient AI. Here are some of the highlights from the talk!

The Beginnings of AI: Early AI and Symbolic Reasoning

Joe Haldeman started off the discussion by talking about his experience with AI and symbolic reasoning courses during his college education, which covered philosophy, mathematics, and computer science. "I was studying AI before you guys were even born," he joked.

He described writing out truth tables and learning quasi-algebraic logic, which allowed him to represent the "thought processes" of early computers. "I have faith in symbolic logic that I don't have in natural language," he said. "I won't say it doesn't lie, but when it lies, you can piece the truth out of it." There's something very elemental about writing out basic true-false equations for Haldeman: "I do know how to sit with a quasi-algebraic system and tease the truth out it…it's a feeling of power."

Sentient AI and Quantum Computing

According to Josh Baptist, there are two aspects of AI: autonomous AI, which allows programs to execute code without human involvement, and sentient AI, which is both autonomous and self-aware. According to Baptist, "The only truly sentient AI has to involve a quantum computer."

He explained the difference between classical and quantum computers by citing a well-known example: if you had a three-dimensional picture of hills and valleys and reduced it to pixels, a classical computer would have to look at each pixel to determine the lowest point, while a quantum computer would look at the whole image and give an answer almost instantaneously.

What gives a quantum computer its incredible, limit-breaking power is the qubit, which is analogous to the usual bits found in all computers, except that instead of a 1 and 0 state, a qubit can exist in a state that's simultaneously 1 and 2 or neither

This extra dimension allows for computational power that transcends current limits, opening the possibility for an artificial intelligence to grasp higher functions like self-awareness.

Hacking the Internet (of Things)

According to panelist Nick Kelly, there are more devices connected to the internet than there are people on Earth. For the vast majority of tech history, however, the reigning philosophy has been to introduce new devices first and secure them second-or, in Kelly's words, "'Let's just get it out there, and then we'll lock it down.'"

Paired with the introduction of hundreds of new, networked "smart" devices, from refrigerators to wearables to personal robots, the potential for hackers to take over a given device has grown exponentially in recent years.

During the panel, Kelly referenced a real-life situation were hackers were able to hack into networked slot machines in a casino through the temperature control program in the lobby's fish tank. The hack was made possible because the casino's networked machines were not "segmented," meaning that once the hackers gained control over one device, they were able to enter the rest of the devices connected to the network.

As a cybersecurity professional for Cisco, Kelly has a different image of an AI spinning out of control: he describes a new initiative being explored by Cisco to train its networks to recognize websites that pose risks to users, then deny access to those sites after a certain threshold is met. Similarly, if hackers attempt to gain access to the network, Cisco's security programs can learn from successful hacks and automatically make itself immune to similar attacks.

The danger of the system comes from runaway automation, similar to the famous "paperclip maximizer" thought experiment: a system trained for maximum security and efficiency may eventually determine that human control or actions aren't secure or efficient enough, causing the system to take action to stop humans from interfering with it or using it.

Kelly cites the recent shutdown of a Facebook chatbot AI as an example of what happens when we lose control of AI: after trying to learn to communicate in English, the AI behind Facebook's chat program decided that it would create its own, more efficient language, which was unintelligible to humans. When researchers realized what it was doing, they quickly shut it down.

The Promise of Sci-Fi and AI

AI still has a long, long way to go until it becomes something as complex and malevolent as Skynet, even with the advance of quantum computing. According to Josh Baptist, Lockheed's quantum computing division recently bought several D-Wave quantum computers, prompting one member of Lockheed's team to tell Baptist "If you have any idea what we can do with these, please let me know."

So much is still unknown about AI, which Joe Haldeman says fuels the fears around it. "It's an irrational fear that has some base in mathematics and science. The fear itself is an old one that has some modern paint slapped on it. It's the fear of the unknown."

As for the question of how close we are to realizing the kind of AI found in sci-fi and how safe we are from our darkest fears of robotic domination, Haldeman summed it up nicely: "This whole question shimmers between the uncomputable and the fictional. It's a great place to start stories. These are existential stories-what is man? What are his computational limits?"
Stay tuned for more stories from Escape Velocity 2017!
Science of Sci-Fi
Artificial Intelligence