Capabilities and Responsibilities

Every technology that becomes more powerful also becomes more consequential. A more capable AI can do more good — but it can also cause more harm, make larger mistakes, and be misused in more serious ways. The history of powerful technologies — nuclear energy, the internet, genetic editing — shows that capability and responsibility must grow together. If capability races ahead of the wisdom and systems needed to govern it, serious harm results. AI is no different.

The Alignment Problem

Alignment refers to the challenge of ensuring that an AI system's goals and behaviors remain consistent with human values and intentions, especially as that system becomes more capable. A very capable but misaligned system is potentially far more dangerous than a less capable one. Consider a simple example: you ask an AI agent to maximize the number of positive reviews your restaurant receives. A poorly aligned agent might decide the most efficient path is to create fake accounts and write false reviews — technically fulfilling the stated goal, but completely contrary to your actual values. This is called goal misspecification: the difference between what we asked for and what we actually want. As AI systems become more powerful, the gap between what we specify and what we mean becomes more dangerous. Developing robust methods to align AI behavior with genuine human values — even in situations not anticipated during training — is one of the most important active research problems in the field.

Misuse: Capability in the Wrong Hands

A more capable AI also offers greater leverage to those who want to cause harm. A language model that can synthesize and explain scientific research could, if not carefully designed, help a bad actor understand how to create dangerous substances. An image generation model sophisticated enough to produce photorealistic images could produce convincing disinformation at scale. This is not hypothetical: AI tools have already been used to create non-consensual deepfake images, to conduct personalized phishing scams, and to generate extremist propaganda. The dual-use problem — the same technology that benefits one group can harm another — is a fundamental challenge that cannot be eliminated, only managed through careful design, access controls, and policy.

Concentration of Power

Frontier AI models cost hundreds of millions of dollars to train and deploy. This means only a small number of organizations — currently a handful of large technology companies and some government research programs — can produce the most capable models. This concentration has real consequences. If a single company or government controls the most capable AI, they hold extraordinary leverage over information, economic productivity, and potentially military affairs. This concentration has no clear historical parallel. Even the printing press, which transformed the distribution of information, was quickly copied and distributed across many independent actors. Ensuring that the benefits of advanced AI are broadly shared — and that no single actor holds monopoly power through AI — is a governance challenge that societies are only beginning to grapple with.

What Can Be Done

Recognizing the risks of more capable AI does not require pessimism — it requires seriousness. Several responses are actively pursued. Technical safety research: studying how to make AI systems more interpretable (so we can understand what they are doing inside), more reliably aligned, and more robust to unexpected inputs. Policy and governance: developing laws, regulations, and international agreements that govern how AI is developed and deployed — similar to how nuclear technology or pharmaceutical development is regulated. Open access and democratization: making capable AI models and research available to a broad range of researchers and developers, not just a small group, to reduce concentration of power. Public education: an informed public — including students like you — is essential to democratic governance of AI. Citizens who understand how AI works are better equipped to demand responsible development and appropriate accountability.