Here we go again: a flurry of Venture Capital money (cf. 1) gets currently invested in the "Next Big Thing", namely - Artificial Intelligence (=AI).
From CB insights (cf. 1), we learnt that 16 start-ups, described in their prospectus as primarily involved in Artificial Intelligence, succeeded in raising funds (series A,B,C) in 2014. Only 2 similar start-ups raised funds in 2010.In the Financial Times (cf. 2), similar results from CrunchBase data were provided. Close to 400 Millions USD have been invested in AI in 2014 (30 Millions in 2010). 103 Millions USD were injected in Sentient Technologies, a financial markets forecasting company that now extends its expertise to medical diagnosis and beyond.
In parallel, the coverage in financial media on AI from Influencers has exploded in 2014. Clearly, it has emerged as a hot topic. Maybe too hot ?I find this subject fascinating however, like probably a lot of us. Therefore, I couldn't resist and I read a lot around.
Artificial Intelligence is indeed a fascinating subject - so fascinating that Science Fiction built many successes around this concept - The list is long from HAL 9000 to Terminator.From this fascination to voluntary or unvoluntary manipulation, there is just a step.
As 2015 Davos attendees discussed actively AI this year (cf. 3), it alerted me that maybe, maybe.. a bubble has already emerged here.
1. Financial Bubbles Matter
Bubbles - everyone knows them nowadays as they have become a predominant feature of our Era: massive flows of money get misallocated. They finance projects based on unrealistic returns expectations - projects that turn deceiving and sometimes even go bust.
Risks of bubbles are here to stay, with the central banks "open-bar" policies (Quantitative Easing).These policies are designed to fight the rampant deflation in the consumer baskets - however they start bubbles everywhere else.
Detecting bubbles has then become the NUMBER ONE priority of investors - in financial markets instruments, real estate, art & unlisted companies. Whatever is outside this consumer basket actually..
Human investors usually do a poor job at discriminating projects, often influenced by the positive sentiment of the crowd. We need new tools.
2. Toward an Early-Warning System on Bubbles
(Un-)fortunately, there is no definitive algo that could allow to detect Bubbles and other forms of extravagant pricing."Laws" in social sciences are not constant over time - unlike in Physics. It's a constant fight in data-mining to extract something relevant.
Few years ago, bubbles could be spotted by a single analysis on the price dynamics. If prices were to shoot-up while accelerating, a bubble could be detected. This is the main conclusion in Sornette 's excellent book.
However with the global & accelerating diffusion of information (Internet), prices tend "to move faster" - they no longer trend linearly therefore. Explosive trends are no longer confined to Bubble cases. They can simply be legitimate repricing.
Conversely, the most efficient markets are now populated by contrarian investors and algos that arbitrate any short-term spike in the price action. Therefore a long-term bubble could develop during a linear trend.. (cf. Apple in 2012).
News Analytics & Media Comments appear like a necessary ingredient to complement any bubble detection - We need to test whether what medias claim to be market drivers, are actually tangible factors.
3. Detecting Bubbles within Unlisted Companies
I already discussed how we could detect bubbles in some private equity investments (cf. 4). The example was based on how Apples overpaid for Beats. It is based on strong correlation between listed & unlisted companies within the same sector. However it implies for some companies of the sector to be listed so that we get publicly available market prices.
In our present case - AI start-ups - I couldn't find any listed company on any stock exchange, so that we could build an "AI index". If someone has any suggestion, I'll be happy to investigate further.
Some might ask: Google maybe ?
Google likes to describe itself as an AI company even if close to 90% of its total revenues are coming from on-line advertising.
I'd answer that there are actually of companies claiming to be involved in AI - 2529 of them (cf. 5).It all depends on the definition we adopt for this fuzzy/general concept of AI.
Anyway, I will concentrate my comments on only one leg of the Bubble Detection Framework: News Analytics & Influencers Monitoring.
4. Why such a buzz in Artificial Intelligence?
I searched into our databases of news to understand how the frenzy around Artificial Intelligence unfolded and I tried to reconstruct the chronology of events.
<Volume of News>
A quick search of the keywords "artificial" & "intelligence" generate a lot of response - we mentioned that already above - the volume is spiking up since 2014.
A lot of comments on this matter, from CEO - top influencers - are reported, (Eric Schmidt, Mark Zuckerberg..etc), from the beginning of 2014.
These quotes on Artificial Intelligence are undeniably very positive.
>> Clearly global tech companies with deep pockets are chasing projects.
Isolated sentences from influencers can attract investors' attention (subliminal message of some sort..), but the buzz really took off with the debate around AI and Increasing Inequality. Hardware AI (robots) can be scary but remote in time, while Soft AI would be more likely to put you soon out of job.
This thesis has turned mainstream with the publication of Brynjolfsson & McAfee 's book: The Second Machine Age. This book written by these 2 economists has been reviewed by leading commentators in financial media.
In 2005, Kurzweil was predicting that AI would surpass any possible form of intelligence in 2045. In 2014, Brynjolfsson & McAfee, after a run in a self-driving Google car, predicted that AI would soon put 80% of us out of job by 2020.
Is there any figure, study to support this thesis ? No apart from this Moore's law - an exponential phenomenon valid since the early stages of computing, that people like to extrapolate. Watch Jeremy Howard (Kaggle)'s TED conference (6). This conference is very interesting on AI but his graph on AI surpassing human (17'50") is definitely too simplistic.
New Technologies (like AI) may not create a lot of job - However it is hardly convincing that it will kill most of the jobs in services. Anyway, with all the debates around rising inequality, the theme - mixing AI and Inequality - quickly became a hit, beginning of 2014.
Soon, a lot of influencers entered the debate, Elon Musk, Vinod Koshla, Peter Thiel, Stephen Hawkins for the most vocal protagonists. Even Bill Gates, very recently.
Elon Musk & Stephen Hawkins are now quoted jointly in this debate, even if their initial comments (Spring 2014) were made around the same time, but independently. Stephen Hawkins signed a paper in The Independent with some other academic researchers, while Elon Musk warned on AI as "our existential threat" on CNBC. Watch the video, it's fun.
When one knows that Elon Musk, Vinod Koshla previously bought stakes in AI start-ups, their comments no longer appear unmotivated. Actually they are (knowingly or unknowingly) stimulating the debate around AI & Inequality to drive up the mark-to-market of their investments. And it worked. It also stimulated a lot of interest from VC funds in various start-ups that admits Artificial Intelligence as a core technology (cf. 5).
5. What's new in Artificial Intelligence ?
Big Data, Statistical Machine Learning & Cloud Computing (in particular with GPU, Graphics Processing Units) were necessary conditions for a significant breakthrough in Artificial Intelligence. But the main trigger for the renewed interest in AI seems to be related to the appearance of a new <KeyWord>: Deep Learning.Artificial Neural Networks are used for 20 years at least, but overfitting them on a set of data is the real issue. Recent progress in Computational NeuroScience (that draw analogies from the brain) highlighted great expectations that one simple & general type of artificial neural networks could be applied to various domains like image processing, speech recognition or tactile inputs.. etc.
6. Google Hiring Spree
Google is fighting hard to remain on top. The company has hired many top-level experts (see below for a (probably incomplete) list). The Same Google bought the London-based DeepMind for 400 Millions USD (from Elon Musk among others / while competing in bidding with Facebook) . When looking closely, DeepMind doesn't have a clear product !! No wonder that investors got exciting by any start-up that claim to be in Artificial Intelligence!
I did a search of <quotes> from the CEOs of these different AI start-ups. What strikes me is how humble and low-profile they currently are with respect to the progress on their current research. We are far from Elon Musk claiming an "existential threat", or the hot debate around the role of AI in increasing inequality.
It seems that people in AI are trying to cool down things recently, with academics opening the debate on the role of Artificial Intelligence, here:http://futureoflife.org/misc/open_letter
11th of January 2015.
Conclusion for Media-Mining & Sentiment Analysis
So what conclusions can we infer from this analysis:
i. Is there any solution for Bubble Detection? Yes, the future here belongs to a combination of AI/Big Data augmented by the human critical mind.
ii. Any framework to detect potential bubbles will most likely combine statistical analyses of the market prices and related metrics on their media coverages.
iii. In situations like this one, where we can't access mark-to-market prices of AI start-ups, we rely solely on News Analytics & Influencers Monitoring to detect potential bubbles. Like any statistical approach, this media-mining technique will take time to uncover the emergence of a bubble AI. Our approach as trend-followers of influencers is too naive when our algos are lagging.These people are influencers and they know it - Creating the buzz around their investments help them inflate the market-to-market of their VC investments.
These algos will need to learn a lesson: Don't Believe the HYPE.
Venture Capital Interest in AI Start-ups
(1) See Bloomberg chart:http://www.bloomberg.com/news/articles/2015-02-03/i-ll-be-back-the-return-of-artificial-intelligence
>> Data Source CB Insights
(2) 20150104 - 1746 - FT - Investor rush to artificial intelligence is real deal (graph) [Intelligence]>> Data Source: FT research, CrunchBase
(3) 20150130 - 1341 - FT - The changing face of employment>> DAVOS about IA & inequality
Bubble Detection in VC
(4) I showed how MarketScience's framework could apply to detect bubbles in the private equity sphere here:http://pentablog.fr/offshore/inauguration-du-chateau-des-hauts-et-20-ans-de-pentalog-france/
The example was based on the growing music streaming industry, and how Apple overpaid for Beats, combining listed & unlisted companies.
(5) Overview of AI businesses
Artificial Intelligence & Rising Inequality
(6) TED conference from Jeremy Howard:
20131227 - 1705 - FT - The robots are coming and will terminate your jobs
20150116 - 1849 - WSJ - 'Soft' Artificial Intelligence Is Suddenly Everywhere
20150112 - 0114 - FT - Scientists and investors warn on AI
20140523 - 2015 - SFGate - Vinod Khosla- Doctors cannot compete with machines [Intelligence]
Steiner C. , How algorithms came to rule our world, Penguin
Erik Brynjolfsson & Andrew McAfee , The Second Machine Age - Work, Progress and Prosperity in a Time of Brilliant Technologies, Norton & Company
Behind any analysis of the impact of AI isRay Kurzweil , The Singularity is Near: When Humans Transcend Biology, Penguin
with Kurzweil's extrapolation of the famous Moore's law. Computing power (and AI)being (super-) exponential with time, one shall reach a singularity (the moment in time when the first derivative becomes infinite) pretty soon.
A significant breakthrough in the analysis of such super-exponential dynamics, and its applications in bubble detection can be found here:
Didier Sornette , Why Stock Market Crashes: Critical Events in Complex Financial Systems, Princeton University Press
Google is hiring in Artificial Intelligence
20150130 - 1442 - FT - Lunch with the FT- Demis Hassabis
20140903 - 1837 - FT - Google hires leading quantum computing expert
20130507 - 1630 - Wired - The Man Behind the Google Brain: Andrew Ng and the Quest for the New AI
Big Data + Smarter Ago = Artificial Intelligence // Recommended reading on the subject
For the anecdote:
20150128 - 0000 - Bloomberg (Mobile) - Texas Hold'em Mastered by Computer With No Wrong Moves
Interesting read for really contrarian minds:20141023 - 1348 - Business Insider - People Are Beginning To Talk Seriously About Google's Search Business Becoming Less Relevant
Computational NeuroScience :