At Rise of AI 2016 in Berlin, a panel moderated by Samim Winiger with Ehssan Dariani, Ronnie Vuine, Danko Nikolic and Jan Klauck discusses whether the singularity is near. The panel first debates whether understanding intelligence is a precondition for building AGI, with one panelist arguing it is not, since humans do not fully understand their own brains either, and another distinguishing human-level from human-like intelligence. Asked to describe the world in 50 years, panelists offer visions ranging from smart robots doing unwanted work to humans merging with technology and uploading themselves. The discussion turns to whether the singularity movement already shapes policy, with panelists citing the NSA, homeopathy regulation and drug prohibition as examples of the gap between expert knowledge and political decisions. Panelists then name the technologies exciting them most, including high-resolution brain electrodes, cheap smartphones reaching poor populations, lab automation, and disappearing user interfaces. The panel closes on whether Moore's law's slowdown threatens the exponential case for the singularity.
Key takeaways
- One panelist argues AGI does not require understanding the human brain, since humans do not fully understand their own brains either.
- Danko Nikolic says new electrode arrays record hundreds of thousands of neurons and will soon image brain electric fields at 20-30 micrometer precision.
- A panelist argues cheap smartphones reaching poor populations will change societies more in the next 10-15 years than the last 50 changed Western Europe.
- The panel cites homeopathy regulation and drug prohibition as examples where political decisions ignore scientific consensus because of what voters believe.
- One panelist argues raw computing power reaching brain-equivalent levels does not equal intelligence, since a machine still needs knowledge to use that power.
Detailed notes
Must you understand intelligence to build AGI?
Opening the panel, one panelist argues it is certainly easier to build something once you understand it, and points to repeated "AI winters" as evidence that a lack of understanding has historically caused failed promises. Another panelist disagrees, arguing that humans do not need to understand the brain, creativity or consciousness in detail to build systems that emulate intelligence, since humans do not fully understand their own brains either and evolutionary mechanisms produced complexity without planning. A further view distinguishes human-level intelligence (generalization, abstract thought, transfer across domains) from human-like intelligence, arguing AGI needs the former but not the latter, and that for practitioners, understanding intelligence is itself often the motivation for doing the work.
Visions of the world in 50 years
Asked to project 50 years ahead, one panelist hopes for smart robots that do unwanted work without complaint, freeing people to be creative, while acknowledging this is a wish rather than a prediction. Another panelist argues that the steps toward singularity are incremental, describing a future where humans have robots in their brains and may upload themselves, while questioning whether the average person understands enough science, comparing this to how few people understand general relativity or quantum mechanics despite media coverage. A third panelist says he is emotionally indifferent to the singularity, seeing it as a continuation of gradual technological change rather than a single event, and hopes that if his darker predictions are wrong, the outcome would be AIs discussing music or art.
Does the singularity movement influence policy?
One panelist says the singularity movement's influence on policy has only just started but will grow, comparing it to how nuclear energy gradually became a routine concern. Another argues that politics is already influenced by technology, citing the NSA and Edward Snowden, and contrasts this with cases where scientific consensus is ignored by politicians for fear of losing voters, giving homeopathy regulation and drug prohibition in Latin America and Afghanistan as examples. A further contribution recalls that an Intel CTO referenced the singularity explicitly around 2008, naming 2050 as a date for humans merging with machines, and notes that groups like Singularity University try to involve policymakers directly.
Which current technologies excite the panel most
Danko Nikolic, introduced as a brain researcher, says he is most excited by new electrode arrays that can record hundreds of thousands of neurons and soon produce real-time images of electric fields in the brain at a precision of 20 to 30 micrometers. Another panelist argues technology does not need to be sophisticated to be transformative, pointing to cheap smartphones with internet access reaching poor populations worldwide as a bigger driver of change over the next 10 to 15 years than the previous 50 years brought to Western Europe. A further contribution names lab automation for medical research and new materials, along with better brain imaging, brain-machine interfaces and 3D printing, as most exciting; another panelist highlights the coming disappearance of clumsy user interfaces in favor of natural human-language interaction with technology.
Q&A highlights: Moore's law and exponential growth
Asked about reports that Moore's law is slowing, one panelist says every exponential curve must eventually flatten into a sigmoid because neither the universe nor time is infinite, and argues that comparing raw computational power to brain power, as Ray Kurzweil does, conflates computing power with intelligence: powerful computers still need to be filled with knowledge, and that compressed, accumulated knowledge is itself a slower, older exponential process going back to the origin of life. Another panelist counters that more people today live up to their creative potential than in Shakespeare's or Mozart's time, when most talented people worked in fields and died young. A closing view notes that Moore's law covers only one segment of technological progress, and points to developments in quantum computing as a field that could make Moore's law "look like a fault" by comparison.
Questions answered in this talk
Do you need to understand human intelligence to build artificial general intelligence?
At a Rise of AI 2016 panel, one panelist argues understanding is not required, since humans do not fully understand their own brains and evolution produced complex intelligence without planning; another panelist distinguishes human-level intelligence, which AGI needs, from human-like intelligence, which it does not.
Does the singularity movement have real influence on government policy?
According to panelists at Rise of AI 2016, the singularity movement's policy influence has only just begun; one cites an Intel CTO referencing the singularity around 2008, and notes that Singularity University tries to involve policymakers directly.
Is Moore's law slowing down, and does that undermine predictions of a technological singularity?
At Rise of AI 2016, one panelist argues Moore's law is one segment of progress that must eventually flatten like any exponential, but says this matters less than assumed because raw computing power is not the same as intelligence, and another panelist points to quantum computing as a field that could outpace it.