AI summary by TwinMind
How AI Could Empower Any Business | Andrew Ng | TED
TED
11m 17s
Historical Literacy Analogy
- Literacy rise few hundred years ago - many thought not everyone needed to read/write
- Society structure: high priests/priestesses/monks read holy books, others listened at temples/churches
- Discovery: much richer society possible when lots of people could read/write
- Today: AI in hands of "high priests" - highly skilled AI engineers in big tech companies
- Most people only access AI built for them by others
Current AI Concentration Problem
- AI projects expensive to build - require dozens of highly skilled engineers
- Cost: millions to tens of millions of dollars per AI system
- Large tech companies (hundreds of millions/billions of users) better at making investments pay off
- One-size-fits-all AI systems work for:
- Web search improvement
- Online shopping product recommendations
- Massive user bases generate massive revenue
Long Tail Problem Outside Tech Sector
- AI recipe fails outside tech/internet sectors
- Hardly any projects apply to 100 million people or generate comparable economics
- Example: Local pizza store owner
- Drives few minutes to buy Hawaiian pizza slice
- Owner has cold pizzas sitting around, different flavors sell out
- Generates data through sales that AI could analyze
- AI could spot patterns: Mediterranean pizzas sell well Friday nights
- Suggest making more Friday afternoon
- Revenue improvement: few thousand dollars/year would be huge deal to owner
Data Requirements Reality
- Hype about AI needing massive datasets
- Reality: AI often works fine on modest amounts of data (single pizza store data sufficient)
- Real problem: small pizza store can't justify cost of hiring AI team
- US: ~500,000 independent restaurants serve tens of millions customers collectively
- Each restaurant different: different menus, customers, sales recording methods
- No one-size-fits-all AI works for all
Small Business AI Applications Vision
T-shirt Company Use Cases
- Demand forecasting: Accountant uses AI to determine funny memes for t-shirts based on social media trends
- Product placement: Store manager takes pictures, AI recommends product placement for improved sales
- Supply chain: AI recommends whether to pay $20/yard for fabric now or keep looking for cheaper options
- Quality control: Quality inspector uses AI to automatically scan fabric pictures for tears/discolorations
Current Reality
- Large tech companies routinely use AI for these problems
- Typical t-shirt companies, auto mechanics, retailers, schools, local farms use AI for zero applications
- Every t-shirt maker sufficiently different - no one-size-fits-all AI works
- Outside internet/tech sectors, even large companies (pharmaceutical, car makers, hospitals) struggle
Long Tail Distribution Analysis
- All current/potential AI projects sorted by decreasing value:
- Most valuable: decides what ads to show on internet
- Second: web search engine
- Third: online shopping product recommendations
- Right side of curve: unique custom projects
- T-shirt product placement
- T-shirt demand forecasting
- Pizzeria demand forecasting
- Millions of projects on tail that no one working on
- Aggregate value of tail projects: massive
Traditional AI Development Barriers
- Past decades: building AI required writing pages of code
- Not everyone has time to learn coding
- Need for new approach to enable broader participation
Emerging Data-Driven AI Platforms
- New AI development platforms shift focus from code-writing to data provision
- Much easier for most people
- Multiple companies working on such platforms
- Example platform concepts using author's team development
Fabric Defect Detection Example
- Inspector uploads fabric pictures to platform
- Shows AI what tears look like by drawing rectangles
- Shows AI what discolorations look like by drawing rectangles
- Green/pink rectangles are data created by inspector
- AI examines data, may need more discoloration examples
- Inspector provides additional pictures to deepen AI understanding
- Timeline: few hours to few days
- Equipment: suitable camera setup
- Result: custom AI system detecting defects/tears/discolorations throughout factory
Broader Applications
- Baker: AI checks quality of cakes
- Organic farmer: checks quality of vegetables
- Furniture maker: checks quality of wood
Current Platform Status
- Platforms need few more years before easy enough for every pizzeria owner
- Many platforms developing, some quite useful for tech-savvy users with bit of training
- Enables empowerment of:
- Every accountant
- Every store manager
- Every buyer
- Every quality inspector
- Goal: build their own AI systems rather than relying on "high priests"
Democratization Vision
- AI creating tremendous wealth, will continue creating wealth
- Only through democratizing AI access can ensure wealth spread across society
- Historical parallel: hundreds of years ago, hardly anyone understood widespread literacy impact
- Today: hardly anyone understands democratizing AI access impact
- Building AI systems currently out of reach for most people
- Coming era: AI will empower everyone to build AI systems for themselves
