Augmented intelligence sees computers helping humans to make better decisions. That will be an opportunity for startups in coming decades as deep-learning apps roll out to mCommerce market, predicts JFDI Guest Mentor Dr. Lawrence Lau.
Augmented intelligence, while still heavily dependent on narrow artificial assistance, becomes an expert in its own right if you aggregate enough of them – much like the The Doctor in Star Trek.
1. Cheap computer power and democratisation of the toolchain
Nvidia, a tech company that manufactures video cards and authoring systems for the gaming market, is developing DIGITS – a platform for people to train neural nets for pattern recognition. This reflects the point that computing power has gotten cheap enough to the point where getting high accuracy (apparently 95% was comparable to humans) is obtainable using a reasonable training regime.
As a case in point, years ago it would have been cheaper to hire a human translator than feed software hundred hours worth of information. But today, universal connectivity would allow inference engines to seamlessly hook smartphone senses like cameras, microphones, and wireless sensors onto cloud servers for the Internet of Things. Previously development tools were only available in academia or industry labs but now widespread platforms allow anyone from Africa to New Zealand to experiment on deep learning.
The recent Media Exploits exhibition in Singapore showed various examples ranging from:
- face recognition such as next generation security systems;
- rich object detection to guide tourists around unfamiliar landmarks;
- age/gender profiling for shopping assistance, to
- video analytics for autonomous vehicles (currently wandering around Singapore’s One-North)
2. Breakout from traditional Artificial Intelligence domains
Historically, research and development into Artificial Intelligence has focused on image recognition or natural language processing with evolution of neural networks that mimic the brain, only constrained by suitable training material. Neural nets learn from mistakes … but much as you teach a child, you need good and bad examples to train and correct the prediction engine. It’s non-trivial. Getting to 90% accuracy is easy for beginners, after that requires tricks of the trade and very good data.
However, the capture of human activity via tweet feeds, link-following, webcams, and communities of interest such as Pinterest and Reddit have created valuable datasets for determining emotions and dislikes (also known as ‘sentiment analysis’), automatic recognition while in motion (video analytics), and data driven recommendation.
Chinese companies are jumping on board, with Baidu and TenCent funding joint projects, as well as taking advantage of low-cost polling to acquire proprietary datasets and similar projects are running at Google and MicroSoft. For example, if I look at a picture say “cat” or “dog”, all a computer sees is a bunch of colors. The best software gets a bunch of expensive experts to do the classification, then mimics them. This is seen typically in optical character recognition, such as the Stanford dataset taking years to clean and test. Acquisition of this capability will allow Chinese firms to escape the competitive “Red Sea of me clones” of Western apps to come up with homegrown innovation. Technology can give firms a competitive edge, much as in 90’s when Google outperformed text-based keyword search with its Support Vector Machine analysis of page ranking, next generation prediction algorithms would be even more sensitive and specific.
This shows how a neural net goes from edges, to features to images to classification e.g dog, cat, etc. (Photo via Coursera)
3. Role of Augmented Intelligence
Generally, ASEAN investors fund operating businesses, not features and are willing to accept market risk with path to revenue based on their assessment of the execution team. Sustaining innovation that create value must address fundamental human needs such as shelter (AirBnB), food (OpenTable) or more hidden desires such as status (Apple). The gap that Augmented Intelligence will address is how humans make decisions, especially in conditions which are highly variable, indeterminable outcomes, and often stressful. In these situations, we revert to gut instinct or more accurately Recognition Primed Decision Making – a human bias towards rejecting options sequentially to go with the first that “works.” How many people do last minute Christmas shopping and rather follow the carefully compiled list, grab a nearest match from the promo-counters conveniently located near registers? Augmented intelligence does the job of overcoming cognitive biases by using subjective expected utility models but their real business is anxiety relief or regret avoidance.
Deep-learning can work, both for good (helping user) or presenting info that triggers (false) recognition. The International Data Corporation predicts that within three years, more than half of developer teams will embed some form of cognitive services within their apps. Since incremental improvement follows more comprehensive feedback learning curve, it will be like an arms race driving uptake of ancillary technologies. Similar to the “browser wars” between Netscape and Internet Explorer in the late 1990’s, which drove the adoption of basic features (embedded movies, scripting, secure sockets etc), cognitive engines will strive to outperform each other in task accuracy and user assistance. As mCommerce acts between the on-offline worlds – intersecting between immersive retail and digital fulfillment– it becomes the battleground, shifting from the attention economy to the experience economy.
While Google has dominated the online search market as ads target the dwindling reservoirs of human patience, the field is still wide open in connecting to a gratifying shopping experience or even professional/personal services. As an endnote to the Media Exploits event at BioPolis, there was a discussion of how Singapore startups can benefit from tapping ASTAR’s expertise with some observations on the failure of imagination, technopreneur/financier risk aversion, and the lack of unique and domain specific training datasets.
This article is based on a post in the JFDI OpenFrog community. Dr. Lawrence Lau is an Intellectual Property Broker under his personal brand Gemwise Invests (“Ideas in the Rough, Roughly the Ideal”).
Edited by Crystal Neri, the Social Media and Content Marketer at JFDI.Asia. Follow her thoughts on Twitter, @nericrystal.