Artificial intelligence (AI) is a catchphrase in startups and venture capital (VC) financing. It’s a new gold rush. In the USA alone, according to the National Venture Capital Association, 1356 AI startups raised a record $18.5 billion in 2019. There has been growth in the number of companies, and funding raised as 2018 witnessed 1,281 companies raising $16.8 billion. However, interest in AI is not new. It started in the 1950s with the idea of machine translation. Like the wave of the ocean, AI activities witnessed ups and downs. Unlike the past waves, AI startups and VC funding focus on innovating products and profiting from them. Each wave of innovation creating success stories of startups relied on a unique technology core. For example, microprocessors and graphical user interfaces fueled PC innovation. Is AI technology core strong enough to create success stories?
What is the target of artificial intelligence? Nobody exactly knows. Human beings have intelligence at multiple levels. In its simplest form, storing and retrieving data is a sign of AI. Once we succeed in imitating such capability, then the goal changes. It seems that the target of AI is moving. As we achieve to implement a certain level of human-like intelligence, we set our goals at a higher level. That may be fine for pursuing curiosity-driven AI research and development activities. But what about the target of developing AI machine capability for the purpose of selling it in the market. Why should customers buy them? Plain and simple economics of consumption determines the decision. The AI capability of the machine should offer them means in getting the job done better, preferably at less cost.
AI technology core forming the foundation for AI startups and VC funding
In the 1950s, raw computing capacity formed the AI technology core. From the 1950s to the 1980s, we added a set of algorithms. Notable ones are, regression analysis, decision tree, support vector machines, clustering, Neural Network, search and optimization, and Fuzzy Logic. Of course, these techniques are useful to detect patterns and gather intelligence in data. The explosion of data at the dawn of the 21st century, particularly produced by social networking, web search engines, and e-commerce platforms, created the demand for applying these AI tools to gather intelligence. Often it was not feasible for human beings to process such data to come to useful understandings. Some examples are shopping behavior or biases of individuals to certain beliefs, events, or issues.
We improved the AI technology core further. The explosion of smartphones has fueled the growth of high-performing low-cost cameras and computing power. In addition to the keyboard and mouse, we have several major sensors to produce real-life data. For example, tiny smartphone cameras can now produce human-eye-like visual data. Compact RADAR, LIDAR, and X-Ray sensors can produce data about the environment that human eyes cannot even provide. The integration of high-performing computational capacity, sensors, massive near real-time and offline data, and computational algorithms has significantly expanded the AI technology core and scope of innovation. Does it mean that we are very much in a position of replacing humans’ cognitive role in understanding situations and making decisions autonomously?
Emerging possibilities of AI innovations—triggering AI startups and VC funding
Of course, there are possibilities of offering better alternatives. For example, according to the US National Highway Traffic Safety report in 2015, 94% of collisions involving automobiles are due to human error. As we are now in a position to collect detailed real-time data surrounding automobiles using sensors like Cameras, LIDAR, RADAR, or ultrasonic, we can use computational algorithms to overcome this limitation. It offers the possibility of removing the cost of the drivers, and also improving safety. For example, research has shown that autonomous vehicles equipped with sensors and cameras can respond to obstacles in about one-third of the time it takes a human driver to react. Moreover, the addition of the RADAR sensor enables autonomous vehicles to see through the fog. Hence, there has been enormous interest in robotic vehicles.
The recent advancement of AI has also triggered the interest in building human-like robots–humanoid. The addition of sensors and AI capability to mechanical outfits appears to be an easy target to develop humans like service providers. Particularly, such humanoid robots could be commercially attractive to offer elderly care services. Moreover, such humanoids will also find the market even in our smart households.
Another area of possibility is imitating the cognitive capability of health service providers. Starting from x-ray, ultrasound to wearable, numerous sensors are producing detailed, and also real-time, vital health data. On the other hand, both the cost and barrier to access to health professionals are increasing. Processing those data in gathering intelligence, and offering advice autonomously appears to be a ripe area of AI innovations.
Indeed, there are limitless opportunities for AI innovations. Subsequently, there has been a surge in both AI startups and funding. But are there issues that pose a threat to trap some of these initiatives in Death Valley?
Challenges of imitating human capability
Human beings have task execution capabilities in three different forms: i. innate, ii. tacit and iii. codified. By born, human beings inherit a set of innate abilities. There are 52 of them in four categories: cognitive, sensory, physical, and psychomotor. To augment them, human beings spend time in education and training to acquire codified knowledge and skill. Moreover, human beings also earn additional knowledge and skills through experience, in tacit form, though. However, education, training, and experience also sharpen innate abilities.
The complexity of automating codified capability seems to be the easiest. Moreover, machines can have far higher codified capabilities than human beings can have. For example, even a simple calculator can compute numbers faster and with higher accuracy than a highly trained mathematician can do. Similarly, a piece of software can handle far larger data volume than human beings can handle.
The second most challenging task of automation is to imitate humans’ tacit capability, earned through experience. In some instances, current generation AI faces extreme difficulty in automating such capacity. For example, a five-year-old can quickly gain experience in how to handle flexible objects. But it’s tough to imitate such capability in machines. However, in some instances, the tacit capacity could be translated into a set of policies, procedures, and standards.
The most difficult part of imitating human-like intelligence has been to automate the innate abilities of humans. For performing even a simple task like serving coffee or having eye contact with pedestrians, we use several innate abilities. Existing AI is extremely weak in automating these innate abilities.
Waves of AI technology advancement and Innovations
Curiosity-driven academic R&D dominated the first wave of AI advancement. Governments of the western world were primarily funding sources. As opposed to billions, the funding amount used to be in tens of thousands of dollars. The 2nd wave of funding came for developing customized applications, mostly for defense and space missions. In contrary to the first two, the current wave of AI is to innovate products. Startups are coming with product ideas, as opposed to customers giving contracts or governments offering R & D grants. Upon sensing commercial prospects, AI funding seems to be coming from every corner of the market. Even Governments, especially in China, are funding companies to innovate products. Corporations, net worth individuals, and national sovereign wealth funds are pumping billions of dollars into startups pursuing AI-related products.
Death Valley to Avoid
Unlike offering something better than having nothing, AI startups are invariably targeting replacing the role of humans. For example, the target is to offer AI innovations for processing medical images. Until and unless AI accuracy is better than human healthcare professionals, there will be no willingness to pay for the product. On the other hand, at a very early stage, primitive medical imaging machines, compared to today’s standard, succeeded in creating a market. Those primitive machines producing noisy, poor resolution images were far better than having nothing. Similarly, unless and until the autonomous driving module is safer than a human driver, customers will not pay for them. Unlike many other innovations, AI products face the human intelligence barrier to overcome for starting to produce revenue.
On the one hand, AI technology is highly illusive. In the beginning, they show rapid growth in imitating human-like intelligence. Unfortunately, before taking over human intelligence, they start showing signs of saturation or oscillation. On the other hand, current AI technology appears to be not strong enough to imitate humans’ innate abilities. Hence, at a later stage, the deadline for release keeps expanding. In addition to burning more money, one after another deadline extension also starts witnessing expire of life previously obtained patents. Such reality has already led to the closure of R&D funding for further advancement of Honda’s humanoid ASIMO. The expected release date of autonomous vehicles from 2019 to 2027 already raises the future of almost $80 billion invested in AI research.
Focus on reality checks
Nevertheless, it does not mean that there is no profitable opportunity for AI innovations. Hence, no way entrepreneurs or VC fund managers should completely shy away. Instead, the focus should be on more reality checks. However, forecasts of massive job loss and the unfolding of the fourth industrial revolution, made by famous think tanks and consultancies, due to AI are fueling irrational exsuberation. The strength of AI and the nature of cognitive complexity being targeted should be taken into consideration to avoid the death valley of AI startups and VC funding. The focus should be on creating early success stories to trigger the flow of waves of AI innovations and start-ups. Otherwise, there is a risk to suffer from AI winter, subsequently stalling the unfolding of the fourth industrial revolution.