Gary Marcus greets me outside the post office on Granville Island in Vancouver, sporting neon-coral sneakers and a blue Arcteryx jacket. I’m in town for a family gathering, and Marcus, who made the move from New York City to Vancouver in 2018 after two decades in the Big Apple, expresses his fondness for the city, referring to it as a paradise. As we embark on our stroll around Granville Island and the nearby Kitsilano beach, our conversation naturally gravitates towards the current landscape of artificial intelligence (AI).
Expressing his disillusionment, Marcus shares, “I’m disheartened by the direction AI has taken. My initial foray into this field did not envision such a massive wealth transfer from artists to large corporations.” Taking a sip of my black dark roast coffee, we continue our discussion.
A distinguished AI researcher and cognitive scientist holding the title of professor emeritus at NYU, Marcus is known for his critical stance on deep learning and AI technologies. Engaging in spirited debates with AI luminaries like Yann LeCun and Geoffrey Hinton, Marcus has established himself as a polarizing figure in the industry. It is during these leisurely walks that Marcus often takes to Twitter to share his thoughts.
This week has seen significant developments in the realm of AI. Google DeepMind unveiled Gemini, its latest advanced AI model with enhanced capabilities in processing vast amounts of video, text, and images. Concurrently, OpenAI introduced Sora, a groundbreaking generative video model capable of transforming brief text descriptions into detailed, high-definition video clips up to a minute long. While AI-generated videos are not novel, Sora’s sophistication has set a new benchmark, flooding my social media feed with awe-inspiring clips. OpenAI touts Sora as a step towards developing universal simulators of the physical world. For a more in-depth exploration of Sora, refer to Will Douglas Heaven’s coverage here.
However, Marcus remains unimpressed. He remarks, “At first glance, the videos may seem impressive, but upon closer inspection, the AI system still lacks a fundamental grasp of common sense.” Pointing out discrepancies such as unrealistic physics and abrupt appearances and disappearances of elements like animals and people, Marcus underscores the shortcomings of generative video technology.
To Marcus, generative video creation exemplifies the exploitative practices prevalent in tech corporations. Numerous artists, writers, and even established entities like The New York Times have raised concerns over AI companies indiscriminately scraping online content to train their models, allegedly infringing on intellectual property rights. Copyright infringement emerges as a primary concern for Marcus. He has advocated for transparent guidelines governing the utilization of copyrighted materials in AI models.
“Utilizing copyrighted content without consent within opaque systems that lack transparency is neither legally nor ethically sound,” Marcus asserts. Pausing at a picturesque vantage point offering panoramic vistas of the city, mountains, and beach, Marcus reflects on the serene setting, a stark contrast to the bustling Silicon Valley. Despite the tranquil surroundings, Marcus often leverages his platform to critique the power dynamics prevalent in Silicon Valley, identifying as an activist in the present context.
When probed about his driving force, Marcus promptly responds, “The custodians of AI exhibit a lack of commitment towards fostering responsible AI, potentially leading to severe societal repercussions.” Having authored a forthcoming book titled Taming Silicon Valley last year, Marcus advocates for stringent regulations and public involvement to steer AI development in a responsible direction.
Highlighting the urgency of action and policy formulation in the AI domain, Marcus warns against repeating the regulatory oversights witnessed in the social media sector. He cautions, “The repercussions of mishandling AI could far surpass the ramifications of the social media debacle.” As we conclude our walk after approximately 12,000 steps, Marcus directs me to a local eatery renowned for its delectable bagels. Indulging in lox with cream cheese under the sunlit sky, we bid adieu.
Later that day, Marcus takes to Twitter to share his assessment of Sora, characterizing it as a remarkable feat akin to morphing and splicing, rather than a gateway to the logical reasoning requisite for Artificial General Intelligence (AGI). He predicts an escalation in systemic glitches as Sora becomes more accessible, cautioning about the challenges that lie ahead.
In a separate segment:
Deeper Learning
A forthcoming satellite project will leverage Google’s AI technology to map methane emissions from space
Scheduled for launch next month, a satellite dedicated to measuring methane emissions will harness Google’s AI algorithms to quantify, map, and mitigate leaks. This initiative, in collaboration with the Environmental Defense Fund, aims to provide detailed insights into methane emissions, aiding in pinpointing high-risk areas and holding accountable the responsible entities.
Shedding light on methane emissions: With methane contributing significantly to global warming, regulatory bodies worldwide are advocating for stringent measures to curb emissions from oil and gas facilities. MethaneSAT, the upcoming satellite mission, will monitor methane plumes emanating from these operations globally. Subsequently, Google and EDF will map these emissions, facilitating research, regulatory efforts, and public awareness. For further details, delve into the coverage by our AI specialist, James O’Donnell, focusing on the intersection of AI with hardware technologies like robotics and chips.
Bits and Bytes
Will AI redefine the concept of videos?
Systems such as OpenAI’s Sora do not capture real footage but instead materialize conceptual ideas. Does the authenticity of these generated videos hold significance? (Source: New Yorker)
Early adopters assess the value proposition of Microsoft’s AI bot
After six months of hands-on experience, testers are divided on the efficacy of the AI bot, raising questions about its utility. (Source: WSJ)
White House earmarks $1.5 billion for a new chip manufacturing facility
As part of the Chips Act, this substantial investment aims to bolster domestic production of critical components, ensuring self-sufficiency in essential technologies. (Source: WP)
AI industry’s exuberance draws parallels to the telecom bubble
When CEOs seek trillions, not billions, in funding, it signals a potential overheating of the sector reminiscent of past tech bubbles. (Source: FT)
