Dr. Eric Haseltine is a neuroscientist and futurist who has applied a brain-centered approach to help organizations in aerospace, entertainment, healthcare, consumer products, and national security transform and innovate. In the first half, he and his wife Dr. Chris Gilbert, a fellow AI researcher, joined Connie Willis (info) to discuss the future of artificial intelligence. The biggest advances in AI these days are coming from the commercial world, through initiatives like Open AI, a consortium that includes Google, Microsoft, and other tech companies. Right now, AI is designed as a crude imitation of our own neural networks, and neuroscience will continue to play a role in AI development, said Haseltine.
In Haseltine's view, the next step in AI is a transition from request-based applications, where the user tells a device what they want, to devices knowing what the user wants before being asked. Knowledge work will increasingly rely on AI, as it saves enormous amounts of research time, he went on. Deepfaked images and video will also get better and better, eventually operating in real time, and will be indistinguishable from the real thing.
One notable aspect of AI's future is that beyond its programming stage, scientists won't fully understand the way it "reasons" in its decision-making. While he acknowledged that fears of AI "going rogue" as a result are valid, Haseltine said that such concerns are misplaced. Ultimately, the human intent behind any application of artificial intelligence is what we need to worry about. "There's never been a technology invented that didn't have the power for good and for bad, and AI's no different," he argued.
Haseltine stayed for calls in the second half. Tom in Missouri wondered how large of a sample a current AI application would need to provide a high-quality recreation of a person's voice. Haseltine replied it would depend on exactly what the final product would be, but that about ten minutes' worth of a recording would suffice. This usually involves a machine-led session where the user follows voice prompts, reading sample passages of text and so on.
Jose, calling from Texas, asked whether AI could ever assist in treating cancer. AI is an excellent way to do so, said Haseltine, as it proposes numerous ways to study individual cases and provide options for treatment. Similarly, Jacob in Kentucky was curious about AI's ability to help people with dyslexia. AI could be taught to see written language the way people with dyslexia do, explained Haseltine, making it a powerful tool to help with reading and comprehension.