Gary Marcus

Simply scaling up large language models is not sufficient for achieving artificial general intelligence. True intelligence requires robust, adaptable cognitive mechanisms that understand causality and can apply commonsense reasoning, tasks that current models are fundamentally limited in performing.

I strongly believe that AI technologies should undergo a public benefit assessment before deployment. The rapid advancement of AI presents profound opportunities, but also significant risks. Without careful evaluation, we risk deploying systems that may cause harm or exacerbate inequality. A public benefit assessment would ensure that AI development aligns with societal values and priorities, prioritize safety and ethical considerations, and foster public trust.

I strongly disagree that we will have a general AI system more capable than any living human by 2028. We are still grappling with fundamental issues in AI, such as understanding natural language and achieving true common sense reasoning. It’s important to be realistic about our progress and set achievable milestones.

I strongly believe that third-party audits should be mandatory for major AI systems. These systems have a significant impact on society, and ensuring their safety, fairness, and transparency is crucial. Independent audits can provide the oversight needed to build trust and prevent potential harms.

I strongly believe that companies should disclose when AI is used in customer service. Transparency is essential for building trust with customers, and people deserve to know when they are interacting with a machine. It also allows for better accountability and helps in setting appropriate expectations.

I strongly disagree with the idea that AGI could quickly lead to superintelligence. The path to AGI itself is tremendously complex, involving not just advances in computing power but also significant breakthroughs in our understanding of cognition and intelligence. Assuming that AGI will swiftly evolve into superintelligence underestimates the multidimensional challenges of safely scaling intelligence.

I strongly disagree that neural interfaces could facilitate easier AI alignment. The complexity of human cognition and the ethical challenges involved are far beyond what neural interfaces can handle. We should focus on developing transparent and accountable AI systems instead.

As a cognitive scientist, I believe it is crucial to regulate AI technologies thoughtfully to protect cultural diversity. AI systems often inadvertently carry biases from their training data, which can lead to the erosion of minority cultures and languages. Without appropriate oversight, AI could homogenize cultural expressions, diminishing the richness of our global heritage. Regulation can help ensure these technologies enhance rather than undermine cultural diversity.

I believe the unchecked proliferation of open-weight models capable of recursive self-improvement is fraught with risks. These models can quickly evolve beyond our control, posing significant ethical and safety concerns. We need stringent regulations and thorough oversight to ensure these technologies are developed responsibly.

I believe AI labs should pause training when AI Safety Institutes detect signs of potential loss of control. Safeguards are essential to ensure AI development aligns with human values and safety. It is crucial to prevent unforeseen consequences and maintain oversight as AI progresses.

As a cognitive scientist, I am deeply concerned about the development of super-persuasive AI that can influence democratic elections. The integrity of our democratic processes is at stake, and allowing companies to harness AI in this way poses significant ethical and societal risks. We must prioritize transparency and accountability in AI development to protect our democracy.

As someone deeply invested in the ethical development of AI, I believe that global AI regulations should indeed include measures to prevent AI technologies from exacerbating socioeconomic inequalities. AI has the potential to both reduce and widen these gaps. Without careful oversight, it may disproportionately benefit those who are already advantaged, at the cost of those who are not. We must ensure that AI advancements contribute to societal betterment broadly, not just for a select few.

Large language models may seem impressive, but they don't genuinely understand the content they process. They lack the common sense and reasoning that true understanding requires.

As a cognitive scientist deeply concerned about the future of AI, I believe that big AI companies must allocate a significant portion of their resources to AI safety. The potential risks of AI are too great to ignore, and ensuring that we develop robust, safe systems is essential for the benefit of society as a whole.

As a cognitive scientist, I believe that revealing AI identities can hinder the authenticity of interactions and may bias people's perceptions. Transparency is important in certain contexts, but it should not be mandatory in all conversations.

As someone deeply involved in AI research, I firmly believe that transparency in disclosing data sources is essential for fostering public trust. When AI systems are open about their data origins, we can better assess biases and ensure accountability, ultimately leading to more reliable and ethical AI applications.

As a cognitive scientist, I recognize the transformative power of AI, which is why it's crucial to establish international treaties to ensure these systems are developed and used responsibly. AI should prioritize the public interest and not be diverted towards enhancing military capabilities. Treaties can help set global standards and encourage collaboration for peaceful advances in AI.

I believe AI will likely lead to a concentration of power because it is being developed and controlled by a small number of large tech companies. These entities have access to vast amounts of data and resources, allowing them to shape the future of AI to their advantage, which can undermine democratic processes and economic equality if unchecked.