by Bill Gates - 2023-12-09 https://www.gatesnotes.com/AI-agents
- AI is about to completely change how you use computers
- A personal assistant for everyone
- Health care
- Education
- Productivity Agents
- Entertainment and shopping
- A shock wave in the tech industry
- The technical challenges
- Privacy and other big questions
Bill Gates' View of AI and Agents
- AI is set to change how we use computers and the software industry
- Software today is "pretty dumb" and requires telling the device which app to use for tasks
- Current software has limited understanding of a user's work, personal life, interests, and relationships
- This will change in the next 5 years with the advent of agents
- Agents are software that responds to natural language and can accomplish various tasks based on the user's life information
- Agents will be able to provide a more personalized experience for users
- The emergence of agents will bring about the biggest revolution in computing since transitioning from typed commands to clicking icons.
Agents vs. Bots:
Agents:
- Dramatically better than past software assistants (e.g., Clippy)
- Capable of nuanced conversations and personalization
- Not limited to simple tasks
- Develop a powerful understanding of the user's activities, relationships, hobbies, preferences, and schedule
- Proactive and capable of making suggestions before asked
- Accomplish tasks across applications
- Improve over time through recognition of intent and patterns in behavior
- Offer deeply personalized services, such as travel planning
Bots:
- Limited to one app
- Step in when a specific word is written or help is requested
- Do not remember the user's activities from one interaction to the next
- Cannot learn the user's preferences or improve over time
Impact of Agents:
- Democratize services that are too expensive for most people, such as health care, education, productivity, entertainment and shopping.
AI in Healthcare:
- Main role: administrative tasks (Abridge, Nuance DAX, Nabla Copilot)
- Future shift: patient assistance and decision making
- Benefits: improved productivity, accessibility for people in poor countries
Health Agents:
- Slower rollout due to life or death consequences
- Evidence needed to prove benefits
- Affordable mental health therapy through AI agents
Unmet Need for Mental Health Care:
- Half of U.S. military veterans don't receive mental health care (RAND study)
- Importance of making therapy more accessible
AI Agents in Mental Health:
- Understanding life history and relationships
- Available when needed, never gets impatient
- Monitor physical responses through smart watches with permission
- Early chatbots: Wysa, Youper.
Exciting Possibilities of Software for Teachers and Students:
- Supplement teachers' jobs by:
- Personalizing learning experience for students
- Liberating teachers from paperwork and other tasks
- Allowing teachers to spend more time on important aspects of their job
Current State of the Art: Khanmigo Bot by Khan Academy:
- Text-based bot that can tutor students in math, science, and humanities
- Explains concepts like quadratic formula
- Creates practice problems for students
- Can also help teachers with tasks like lesson planning
Future Potential of Agents in Education:
- Agents will open up new learning opportunities
- Supplemental instruction for students who cannot afford private tutors
- Customized to individual student's interests and needs
- Richer and more personalized experience than text-based tutors
- Competition in productivity field: Microsoft (Copilot), Google (Assistant, Bard)
- Capabilities: turn written documents into slide decks, answer natural language questions about spreadsheets, summarize email threads
- Future functions: help create business plans and presentations, generate product images, attend meetings and answer questions
- Integration with daily life: order flowers for friends, schedule catch-ups with old acquaintances
- Assistance in various settings: office or no office
- Personalized assistance similar to executive personal assistants.
AI in Entertainment and Shopping
Advancements:
- Already assists with TV selection, movie recommendations, books, shows, podcasts (Pix)
- Spotify's AI-powered DJ makes personalized music suggestions and calls users by name
- Agents help act on recommendations: placing orders or signing up for streaming services
- Customized news and entertainment based on interests (CurioAI)
Examples:
- Recommends movies, shows, books, podcasts based on past preferences
- Helps with shopping decisions like camera purchase
- Offers to sign up for streaming services if needed
- Makes suggestions even when users are unsure what they want.
Impact of Agents in Tech Industry
Agents as Next Platform:
- Agents will help with various activities and areas of life
- Ramifications for software business and society will be profound
- Agents will be the next platform, replacing current ones like Android, iOS, and Windows
Creating New Apps/Services:
- No need to know code or graphic design to create new apps/services
- Users can simply tell agents what they want
- Agents will write code, design app, create logo, and publish it online
Impact on Software Industry:
- Agents will affect how software is used and written
- They'll replace search sites, e-commerce platforms, word processors, spreadsheets, etc.
- Businesses in different areas (search, advertising, shopping, productivity) will become one business
Competition and Pricing:
- Many AI engines will be available, leading to competition
- Agents may be free with ads or paid for premium services
- Companies will have incentive to make agents work on users' behalf
Challenges and Ethical Considerations:
- Need to address questions about technology and its use
- Previous writings on ethical considerations of AI apply here.
Technical Challenges of Personal Agents
Data Structure:
- No standard data structure for agents yet
- Need a new type of database to capture nuances of interests and relationships
- Maintain privacy while quickly recalling information
- Vector databases may be better for storing machine learning model data
Number of Agents:
- Will there be separate agents for different aspects (e.g., personal, therapy, math tutor)
- Timing of collaboration vs. separation
Interaction Methods:
- Apps, glasses, pendants, pins, holograms are possibilities
- Earbuds may be the first big breakthrough
- Agents can speak or appear on phone if needed
- Enhance sound by blocking out background noise and amplifying hard-to-hear speech
Challenges:
- Lack of standard protocol for agent communication
- Affordability to make agents accessible to everyone
- Easier ways to prompt the agent for accurate responses
- Prevent hallucinations, especially in sensitive areas like health
- Ensure agents don't harm people due to biases or malign purposes
Privacy and Other Big Questions
- Issues of online privacy and security will become more urgent
- Need to decide what information agent has access to
- Confident data is shared only with chosen people/companies
- Ownership of data: who owns the shared data, how it's used appropriately?
- Advertising concerns: no one wants targeted ads based on sensitive information
- Law enforcement use of agents: can they be used as evidence against individuals?
- When will agent refuse harmful actions?
- Values built into agents: who picks them?
Sharing Information with Agents
- Need to control what information shared with friends/family
- Personal information vs. professional information: when should agent keep quiet?
- Agents may affect how we interact with loved ones
Impact of Agents on Society
- Agents could make everyone have a high quality of life without working much
- Raises questions about purpose in life and societal function
- Many people still want to get an education despite agents having answers
- Potential for safe and thriving society with lots of free time remains uncertain.