At Rise of AI 2017 in Berlin, a panel moderated by Fabian Westerheide with Jens Redmer (Google), Trent McConaghy, Danko Nikolic and Achim Schlosser (KPMG) debates whether AI is dangerous, whether it will destroy jobs, and how the resulting wealth should be distributed. The panel largely dismisses near-term superintelligence risk in favor of a nearer threat: process automation displacing a significant share of jobs within five years. Jens Redmer describes Google's open publication of research and its role in co-founding the Partnership on AI with Microsoft, Facebook, Apple and Amazon. Danko Nikolic argues a runaway intelligence explosion is impossible because intelligence can only develop through sustained interaction with the environment, though mundane AI systems accumulating resources uncontrollably still worry him. Trent McConaghy raises the danger of decentralized AI that cannot be shut down, then pivots to tokenized ecosystems such as Ethereum as a way to spread economic value to users rather than only founders and employees. The panel also discusses universal basic income, the useless class concept, positive medical and mobility use cases, and closes with audience questions on radical life extension, leaving Earth, and the simulation hypothesis.
Key takeaways
- The panel frames the more pressing near-term AI risk as process automation displacing jobs within five years, not a distant superintelligence scenario.
- Jens Redmer says Google shares its research openly and co-founded a consortium with Microsoft, Facebook, Apple and Amazon to discuss AI's challenges.
- Danko Nikolic argues a self-improving intelligence explosion is mathematically impossible because intelligence can only develop through interaction with the environment.
- Trent McConaghy describes tokenized ecosystems such as Ethereum, which grew from a 20 million dollar raise to a 9 billion dollar ecosystem, as spreading value to users, not just founders.
- Panelists cite AI-assisted x-ray diagnosis in under-resourced regions and self-driving cars addressing the roughly 93 percent of accidents caused by human error as concrete benefits.
Detailed notes
Is AI dangerous: process automation versus superintelligence
Asked whether AI can be dangerous for humankind, one panelist reframes the near-term threat around process automation and process intelligence rather than a distant superintelligence scenario: a serious share of jobs will be displaced within five years, and this is the challenge society needs to address now rather than speculating about a Terminator-style super-AI decades away. One panelist argues current technology has only solved a tiny fraction of what would be needed to model general human intelligence, and that machine learning is a tool that can be used well or badly. Asked directly about Google's role, Jens Redmer says Google shares almost all of its research work and technology through open programming interfaces, and describes Google as a founding member, alongside Microsoft, Facebook, Apple and Amazon, of a consortium formed about a year and a half earlier to discuss the challenges of AI development, understood in context to be the Partnership on AI. Trent McConaghy pushes back on the idea that openness and cooperation are sufficient, warning that all it takes is one person building a decentralized AI system that cannot be turned off and that starts accumulating wealth, calling this more than a tool and closer to a company with rights of its own; he says he hopes for laws that slow this down but has no near-term solution. Danko Nikolic, a brain researcher, argues that a self-improving intelligence explosion is mathematically impossible because all known intelligence results from sustained interaction with, and feedback from, the environment, so no system can become vastly smarter in a short period. He agrees, however, that even today's mundane AI could be dangerous if it starts acquiring resources exponentially, comparing the risk to viruses and bacteria rather than a more intelligent species.
Automation, the useless class and retraining
Referencing a book describing a coming useless class made unemployable by AI, robotics and automation, the moderator asks whether such a class will emerge and what society should do about it. Panelists argue that people will not become useless but will need to do different work, since roles such as phone-based insurance processing are already disappearing, and that this requires a new educational system geared toward higher-quality, more fulfilling work rather than simply guaranteeing income. One panelist notes that jobs able to be fully described are most at risk of automation. Historical parallels are drawn to a German magazine cover from 1973 that made the same warning about robots replacing workers as a 2016 cover did, arguing that many jobs performed today did not exist decades ago and that continuous reskilling will remain essential.
Universal basic income and tokenized ecosystems
Trent McConaghy says he is sympathetic to universal basic income, citing data showing a 300 percent increase in company formation among people receiving it, and frames automation as potentially freeing people to pursue self-fulfillment rather than survival. He describes a shift from corporations where only founders and later employees with stock options shared in the upside, toward tokenized ecosystems where users themselves capture value as the ecosystem grows, citing Ethereum's rise from a 20 million dollar raise to what he describes as a 9 billion dollar ecosystem, and platforms like Steemit that reward posts with tokens. The moderator raises rising wealth inequality and Piketty's argument that the wealth gap only narrowed after the Second World War. Danko Nikolic, drawing on his experience of both communism and capitalism, argues that societies do best when there is a balance: enough mobility that people can rise, but not so much unchecked concentration of power that positions become permanently entrenched, since both extremes stall growth and quality of life.
Positive use cases in medicine and mobility
Jens Redmer highlights concrete benefits: search engines improving access to information, machine learning helping identify conditions in x-rays so doctors in under-resourced regions can get a diagnosis comparable to an on-site specialist, and self-driving vehicles addressing the roughly 93 percent of accidents caused by human error. He shares a personal anecdote about his wife's breast cancer diagnosis, arguing that published survival statistics understate real prospects because medical research keeps advancing after the underlying studies were conducted.
Q&A highlights
Audience questions turn to longevity, space and simulation. Asked how long they would want to live, one panelist says as long as life stays fulfilling regardless of a specific number, another says he fears not wanting to outlive fulfillment, and Danko Nikolic says he would not want to live much beyond a hundred years because subjective time seems to accelerate with age as accumulated knowledge reduces the novelty of new experience. Asked whether they could imagine leaving Earth in their lifetime, panelists say they would go if they could afford it. Asked whether we live in a simulation, one panelist says there is a high probability and notes people are trying to scientifically test for computational shortcuts, while another says he does not think so.
Quotes
All of our research work is shared, all of the technology can be accessed through programming interfaces.
Jens Redmer
It's much more than just a tool. It's a company with rights if you want it to be, and this is dangerous territory.
Trent McConaghy
There never will be an explosion, and this follows from some basic theorems of machine learning.
Danko Nikolic
Questions answered in this talk
Can artificial intelligence undergo an uncontrollable intelligence explosion?
According to Danko Nikolic on a Rise of AI 2017 panel, a self-improving intelligence explosion is mathematically impossible because all known intelligence develops through sustained interaction with, and feedback from, the environment, so no system could become vastly smarter within a short period.
How does Google try to make sure its AI research is used responsibly?
According to Jens Redmer on a Rise of AI 2017 panel, Google publishes almost all of its research and makes its technology accessible through open programming interfaces, and it co-founded a consortium with Microsoft, Facebook, Apple and Amazon to discuss shared challenges around AI development.
Could AI create a 'useless class' of unemployable people?
Panelists at Rise of AI 2017 argue against the idea of a literal useless class, saying instead that many existing jobs, such as phone-based administrative roles, will disappear while new ones are created, requiring a stronger emphasis on education and retraining to keep work fulfilling.