This August, an article headlined “How to convince Venture Capitalists you’re an expert in Artificial Intelligence” went viral after outlining strategies to bluff investors hungry for AI. Suggestions included: “propose your own ‘Net’”, “reference a paper on Arxiv as if everyone has read it”, and “sign up for the Google TPU Alpha”.
McKinsey & Company estimates total 2016 external investment in AI at US$8 – $12 billion. The boom has seen countless companies scrambling to jump on the bandwagon — even if their products have little to do with AI.
“In Silicon Valley, it is common for venture capitalists to encounter fake AI startups,” says Sue Xu, managing partner at early-stage fund Amino Capital. “Some people are smart at making up stories. When startups have no market traction, they will instead add hype to drive attention from venture capitalists. And the recent hype is AI.”
The trend can be seen on social media. “Inspirobot” for example bills itself as “an artificial intelligence dedicated to generating unlimited amounts of unique inspirational quotes.” Actually, the popular bot simply grabs a background image and stuffs text fields with randomly generated and often nonsensical phrases.
Some chatbots are also what we might term fake AI. A common characteristic of such products is the co-opting of APIs (Application Programming Interfaces) and SDKs (Software Development Kits) from open-source libraries. API.ai is a company that helps developers build conversational bots. Last year, the company was acquired by Google. Since then, many chatbot companies have used their open-sourced codes without attribution.
Financial technology is another area likely to gestate fake AI products, says Idris Mootee. The CEO of global innovation, design and consulting firm Idea Couture, Mootee believes many fin-tech companies can mistake computer-assisted approaches for real AI: “High frequency, high speed automated trading, for example, is not necessarily AI powered.”
The difference between fake AI and real AI
Generally speaking, AI refers to intelligent machines that can learn, improvise and evolve like human beings. However, AI remains a loosely-defined and oft-misunderstood term, applying to different machines with different degrees of intelligence. So, how to spot fake AI?
Data and machine learning models are key and complementary elements in an AI solution. Real AI startups will have unique and specialized datasets such as aerial drone images of construction sites or traffic condition reports from self-driving cars.
Most fake AI companies however ignore the importance of datasets and instead simply apply automation to their products. But automated machines require instructions, while intelligent machines don’t because they can learn from data. “There is a fine line between automation and AI,” says Mootee.
Take the example of a delivery system. An automated Amazon delivery robot follows a specified route. But an AI-enabled delivery robot can learn how to navigate through crowded streets and even change routes due to its visual processing technology and machine learning algorithms. When more data is collected and algorithms are optimized, AI-enabled delivery robots have the advantage of being able to deal with unexpected situations.
The path to real AI
“When I read about AI, 80% of the time it’s just flat out wrong information,” says Prof. Stewart Russell of the University of California at Berkeley. This misunderstanding of what AI really is has largely enabled the emergence of fake AI.
This doesn’t mean investors should shy away from AI startups. Regardless of the hype and the fakes, real AI remains as promising a 5-10 year prospect as the digital transformation of the 2000s or the mobile internet explosion of the 2010s.
To protect themselves from fake AI, Mootee advises companies to carefully consider what problems they are trying to solve and how and why they might benefit from AI-enabled solutions before blindly jumping into AI.
Real AI startups prioritize the generation, collection and processing of both structured data and unstructured data such as text message or social media comments. Says Mootee: “With proper data, you can do experiments, and you can output something meaningful and build your AI base around that. Then you are really utilizing AI, you are not cheating yourself.”
Of course, along with robust data, a real AI company also needs AI scientists. If a startup has no one experienced in AI R&D, data processing, product or business development, it can hardly be expected to come up with any real AI products.
Fake AI is an aggregation of unethical marketing strategies and immature tech solutions. However, because fake AI does not offer a true competitive advantage, and the market is becoming increasingly aware of what real AI can achieve, fake AI products and novelties like Inspirobot are unlikely to stand the test of time.
Journalist: Tony Peng | Editor: Michael Sarazen
Where is Inspirobot source code? Is it just Mad Kib style XML tags? Can it be upgraded to true AI learning system? Would it get less hilarious and creepy, or more??
Pingback: Implementing A Solid AI Strategy | Global Research Syndicate
Pingback: Implementing A Solid AI Strategy | Research Snappy