How is generative AI changing the world?
Generative AI makes language-model capabilities widely available. This raises questions about how economic structures and social contracts may change.
Questions
Pursuing questions that matter
Research begins with questions about what we know and what remains unclear. The questions below organize my research agenda in the context of generative AI.
01
AI and society
Why did AI suddenly advance so much?
Transformer architectures, large-scale computing and internet-scale training data produced capabilities that surprised many experts. Studying that history may help explain what comes next.
What are the risks of rapidly advancing AI?
AI agents may spread misinformation or make critical systems more brittle. We need to test whether current safety methods can govern systems deployed at this scale.
What will our world be like when AI becomes the main authority of knowledge?
More people now ask language models for answers instead of consulting primary sources. How will societies verify claims and establish shared knowledge if that habit becomes dominant?
02
AI and education
What does generative AI mean for higher education?
AI systems can produce essays, code and analyses, making it harder for universities to assess student learning. Should courses place more weight on judgment and systems thinking than on information retrieval?
How is AI going to change our schools?
Adaptive AI tutors could make instruction more individualized than the standard classroom model. This raises practical questions about how schools should change their teaching and operations.
What does the teacher of the future look like?
If AI handles more instruction and explanation, teachers may spend less time presenting content and more time designing learning, mentoring students and teaching critical evaluation.
What should students learn from AIs, and what should they learn from humans?
Schools need to decide which forms of knowledge AI can teach well and where human judgment, context and empathy still matter. That choice will shape future curricula.
03
Organizations and professional work
How does AI impact professional work?
As AI handles more routine cognitive work, judgment and ethical responsibility may become more valuable. Organizations will need to reconsider how they train and evaluate professionals.
How does AI blur the boundaries between software and management?
If software organizes workflows, coordinates teams and revises plans, the difference between a tool and a manager becomes less clear. How should organizations assign authority in response?
What does a science of human organizations involving AI agents look like?
Most theories of organizations assume that all participants are human. We need new research to understand how people and AI agents coordinate and how that affects performance.
What will our world and organizations look like when robots can have LLM brains?
Connecting language models to robots brings AI decision-making into physical spaces. How will organizations use these systems, and which decisions should remain under human control?
04
AI agents and entrepreneurship
Can AI agents be autonomous entrepreneurs?
AI agents can use software tools, search for opportunities and launch small ventures. Does that make them entrepreneurs, or does entrepreneurship still require human agency?
Can entire business organizations be run by AI agents alone?
A firm without employees would require AI agents to manage operations, customers and changing market conditions. Research can test where such organizations work and where human oversight remains necessary.
Can AI agents identify good opportunities?
Language models can find patterns in large datasets, but it is unclear whether they can recognize genuinely new opportunities rather than repeat patterns from the past.
Do AI agents make good business decisions?
AI systems can synthesize complex information and still make confident errors. We need evidence about when their business judgments are sound and when people should override them.
05
Human–AI judgment
How does AI influence human decisions?
Fluent AI advice can be persuasive even when it is biased or wrong. Studying that influence can show when AI supports judgment and when it weakens independent thinking.
What are the best ways to work with AI?
AI can search broadly while people contribute context, intuition and responsibility. The practical question is how to divide work so each contributes what it does best.
When do human–AI hybrid arrangements work better than either humans or AI alone?
Combining people and AI does not always improve results. We need to identify the tasks where human oversight adds measurable value and those where it only adds delay.
When should humans be in the loop?
Organizations must decide which decisions are too consequential to automate. The answer depends on moral, financial and social stakes, as well as who remains accountable.
06
LLM cognition and identity
How do large language models actually think?
The internal workings of large language models remain difficult to interpret. Better explanations may help us predict their responses and understand the limits of their reasoning.
Are large language models a new form of life?
Language models can sustain conversations and adapt their responses, but they are not biological organisms. What concepts should we use to describe and relate to them?
Are large language models self-conscious?
There is no scientific consensus that current AI systems are conscious. Even so, their ability to sound self-aware affects how people relate to them and how designers should present them.
What values do LLMs hold?
Model outputs reflect training data and the choices made during development. Studying them can reveal which values and biases these systems reproduce.
How do LLMs make managerial and business judgments?
When language models allocate resources or suggest strategy, they rely on patterns learned from training data. Their recommendations may contain hidden assumptions about business and economics.
Do LLMs have feelings?
Language models can sound empathetic even if they do not feel emotion. That performance can still create real attachment, with consequences for users and designers.
How do LLMs view humans?
A language model’s representation of people comes from its training data. Examining that representation can reveal the assumptions and biases it projects onto users.
07
Explored in research
How can digital marketplaces be deliberately designed as innovation systems rather than just allocation mechanisms?
Generativity theory asks how platforms enable people to create things their designers did not predict. The challenge is to leave room for those contributions while maintaining rules that sustain the platform.
What role does random experimentation play in achieving exceptional entrepreneurial outcomes?
Generative models make ideas and prototypes cheaper to produce. This may reward founders who run more varied experiments, but we do not yet know when randomness improves the odds of an exceptional result.
How do new forms of digital organization defy traditional knowledge constraints?
Crowdsourcing and marketplaces helped firms reach knowledge beyond their boundaries. Large language models offer another route, raising the question of whether AI can solve search and coordination problems better than human crowds.
How do cognitive biases affect the timing and pricing of micro-acquisitions?
AI coding tools have made small software ventures easier to create and sell. That gives founders more acquisition decisions and makes it important to study how effort and bias affect price.