Fireside Chat, Interview
a16z Podcast | Machine Intelligence, from University to Industry
- Deep learning is forecast to surpass trained radiologists in accuracy within approximately five years, potentially rendering current training obsolete, while the FDA is expected to eventually adapt to these technologies despite current "black box" regulatory hurdles.
- Future AI development may lead to systems vastly more intelligent than humans, creating an intelligence gap comparable to that between humans and ants, which necessitates intentional ethical legislation to prevent dangerous outcomes, particularly given the risk of military entities like North Korea ignoring rational constraints.
- Gaming companies like Electronic Arts are positioned to succeed in AI due to superior risk tolerance and business models, whereas C-suite executives are generally considered too risk-averse to commit to the multi-generational R&D required for difficult AI breakthroughs.
- Google faces a strategic risk where disruption to its advertising revenue would require a new major cash source to sustain reinvestment, and the company is expected to need a long time to achieve a workforce where 25% of software developers write deep learning code.
- Innovation in large public companies is expected to stagnate without curiosity-driven R&D, and a major cancer therapy player with 60% market share may need to destroy its existing business model to succeed, though internal resistance to this risk is anticipated.
- Significant industry churn is predicted to remove all current M&A participants from the market within five years, while software capabilities are expected to evolve toward self-improving systems and a shift from big back-end processing to mobile applications.
- The next major battleground for AI is anticipated to be StarCraft, with predictions of a future human versus Google AI match, and it is estimated that roughly 45% of the research behind AlphaGo originated from the University of Alberta.
- Most doctors are expected to accept AI for human augmentation provided the focus remains on patient care, while the broader societal integration of computers as peers rather than tools is predicted to occur beyond the current speaker's lifetime.
- Long-term progression of increasingly intelligent AI could result in creating a "god" entity that must be ethically sound, and without intentional governance, there is a significant risk of creating dangerous systems that machines may not treat humans as peers.