Fundnel Spotlight • 21 August 2020
Drilling deeper into the exponential growth of AI, we spoke with an experienced field of international experts across the spectrum of investments, innovations, and operations.
Building on the explosive development of AI innovations and its consequences, artificial intelligence is touted to be embedded in almost all tech solutions today. Has it really become an essential weapon for a company to triumph over its peers? Or is it thrown around mindlessly because it is one of the best ways to convince the VCs that a business can achieve scale while staying lean?
Fundnel’s Head of Investments Development, Carlos Camacho recently moderated a panel discussion with four distinguished guests who are leading breakthroughs in artificial intelligence and the use of data.
Catch a recap of the discussion above (kicking off around the 5-minute mark) or read on for the highlights from the conversation below.
Where do you draw the line in defining a true AI solution (given that almost all companies are touting AI in their tech these days)?
Marie: I believe that there is no true or fake AI. I think our perception of AI has changed over the years, and I think today it's more about the smart assembling of different blocks of technology rather than creating your own AI that determines the quality of a solution.
Terry: Here at AIC, we are a deep technology, AI-focused fund, but we also recognise that one does not need to be an AI company in order to be a successful company — it is just how we narrow down our targets here at AIC. Under our concept, companies have to have a full data infrastructure to capture all data from end to end to lend them a path to evolve their algorithm over time and to continue to improve upon it over time with machine learning.
Marco: You may have the best AI in the world but it will not work if it doesn’t solve a problem.
Our company, CHRONOS, doesn't have AI in itself, but our technology uses AI as a situational awareness solution to track movements of everything under our mesh network, without the use of the internet, cell towers and GPS satellites. We run six algorithms at all times to be able to have a point of history and sensor data.
RJ: I see companies calling themselves AI companies all the time, and it has become a marketing buzzword. Many companies use AI to categorise themselves when all they're doing is using platforms like Tableau to visualise and analyse data. As a common example, I've seen companies use machine learning for use cases where traditional statistical methods — like Bayesian statistics — can achieve the same results. I believe companies that are not using supervised and unsupervised deep learning, pattern recognition and complex neural networks are not true AI-based solutions.
A true AI company can solve problems that were formerly impossible for humans to solve or too time consuming or labor-intensive, or that require complex human intelligence to solve. The most spectacular example is self-driving cars.
At present, Price.com uses AI for our product matching engine that matches broad structured and unstructured products across all retail formats — new, used, refurbished, rental, private label and offline — to offer consumers advanced comparison shopping. We also use AI for our web-scanning technology, for categorisation and normalisation of a catalogue of over one billion products.
We have plans to expand our AI capabilities by incorporating the latest improvements in deep learning to further enhance the existing current matching technology and provisional search as well.
Terry, how do you evaluate AI startups and separate the bad from the good from the great?
Terry: Leading from what RJ has talked about when he was trying to classify a true AI company, he's actually mentioned a lot of what a great AI company would possess. Being able to crunch structured and unstructured data with a full data infrastructure is what makes a great AI company.
At AIC, we love companies that have a full-stack technology solution, and a good analogy for that is if you're a wine maker making a bottle of wine, a full-stack solution would mean being able to control the entire process — from planting the grapes all the way to production. And if something goes wrong in the process, you could actually go all the way back to one part of the production, even all the way back to the grape or soil to tweak the step in that process in order to improve the entire product.
That’s what we're really excited to see in companies and is what’s present in some in our portfolio companies — great deeptech solutions.
Marco, RJ, what are the biggest hurdles you face, as an AI company, when it comes to fundraising?
Marco: When you innovate, you are not in the market in a way — you are creating a market. On the other hand, investors are mostly interested in companies that have already sold products. Hence, access to the capital needed to bring new intelligence to market is heavily restricted. This leads to the second hurdle of not being in control of a fundraising or delivery timeline.
Next — competency. A lot of investors we meet have a limited understanding of AI and are unable to see the global impact of the new technology in front of them.
Lastly, there is no competition today, which we wish to have, because it will confirm the market and gives us a valuation.
RJ: I believe investors need to have a basic understanding of the talent required, engineering challenges, and turnaround time to implement or even improve AI-based solutions.
I feel the biggest questions are usually based around misconceptions and how long it takes to make even a marginal improvement to AI-based models / technologies.
Another big factor is — when every company is labelling themselves as an AI company, it is very hard to distinguish between what is actually an authentic AI-based technology versus who is just using it as a marketing tool for fundraising. This can lead to true AI-based solutions being undervalued for the innovative or even groundbreaking technology they're developing.
Marie, what are your thoughts on the deployment of AI solutions across companies in Asia? Where have you seen AI deployed, most commonly, in either company types of divisions – and are they usually developed in-house, with partners, or acquired?
Marie: Some business leaders tend to see AI as a magic wand, based on use cases they’ve read about. Once they realise that it's not a plug-and-play approach to implementing AI, they find it too complex and abandon the thought.
Other leaders have the misconception that they need to hire the top AI talent to implement AI-based solutions, while others just don’t believe that the solutions will be implemented at the right time because AI is such a rapidly-evolving space.
I would say it's always the right time to consider deploying AI in a company, and whether that is done in-house or outsourced, I believe it should be a mix of both — outsourced to bring part of the AI expertise into a company and in-house to structure, organise and manage your own data within the organisation.
The common places where AI is applied today is in security, resource optimisation in predictive maintenance or industrial companies, and personalisation and targeting, much like what RJ is doing with Price.com. Another common application of AI is in the automation of rapid intelligent tasks such as fraud detection in banking — not an easy task but very repetitive.
Terry, in which sectors and subsectors do you see as the most attractive AI opportunities? And what does the split between consumer, enterprise and gov-tech look like to you?
Terry: Sectors that are best positioned to utilise machine learning and AI are sectors where every step of the process is measurable, e.g. translating physical properties into measurable data points. Cybersecurity is a prime example, where information enters the system, and is trackable.
On the split between enterprise and consumer, I think most AI companies today would be B2B focused — this is what we’re seeing in our portfolio. We’ve invested slightly over USD70 million over 16 companies globally, and 90% of our companies today are B2B focused. For example, SenseTime, one of the largest unicorns in China — whose last round came in at USD7 billion — are entirely B2B, serving large corporations like Alibaba, the Chinese government for smart cities, etc.
Having said that, there will be B2B elements that lends itself to B2B2C applications. For example, SenseTime uses its computer vision and visual recognition AI to power Meitu XiuXiu— a beautifying app in China extremely popular with consumers.
We think the nature of AI is such that only enterprises today would have a larger budget to incentivise AI innovation and growth by rewarding AI companies to innovate and continue their R&D development, and I want to throw out a statistic out there for everyone to think about, which came as a surprise to me:
What do you think was the average AI budget companies allocated in 2019, before Covid?
Carlos: I'll go with maybe six figures. I think Marie is probably going to get the closest to this. What do you think, Marie?
Marie: I would also say around six figures.
Terry: You’re both very correct, but it's in the low six figure range. It was only around USD100,000 in 2019, and I'm hoping to see a very different number in 2020 post-Covid, but the fact is that these are not evenly distributed amounts. Only really large corporates will have million dollar budgets to throw at really good AI companies to innovate on. So my short answer is —
B2B will prevail in the coming few years before we see a shift towards consumer AI.
Marco, what is your take on public-private partnerships, as I know CHRONOS is in discussions with a number of governments around the world. How are governments looking at AI solutions, and what types of solutions are they most interested in?
Marco: I think the interest is multiple for CHRONOS. As a situation awareness solution, the data is created by the location of our electronic devices.
One of the use cases that we're working on developing is for Covid, where you would have our mesh network in a city, building or port equipped with thermal cameras that can locate somebody with fever. We would be able to create a digital ID of this person based on their movement and cell phone signal, and at the same time, get the history of their proximity to others and a prediction of the person’s journey from that point. This would be extremely helpful for hospitals or building owners who want to track potential cases.
We have another proposed use case for Sentosa, where we can cover the entire island with our mesh network and provide staff with ID cards. This will allow us to measure the productivity of the staff and ensure that they are at the right locations at the right time, but most importantly for the visitors to see the heat maps for crowd control and social distancing purposes.
The third use case might be in public security, where issues such as terrorism and national security are concerned.
Marie, do you believe that leaders and technologists are aligned internally and what do you think they’re typically wary of and hopeful for?
Marie: They are not aligned, and they should not be aligned because they are pursuing different goals, so while it's not a problem at the start, they will need to work on that.
What happens very often between leaders and technologists is that there is a lack of communication. I believe that technologists, as experts, need to recognise how to translate complexity into simple, understandable terms for business leaders to understand. On the other hand, business leaders should also make an effort and take the time to acknowledge how much they understand about AI, and to be open to learning about the complexities with regard to corporate innovation.
With that, both parties would be able to align on the expected risks, evaluate them together, identify the possible roadblocks, and turn assets into resources. Of course, there are instances where technologists and business leaders live on different time scales, and as such, alignment on project preparation and management may also need to be discussed.
RJ, antitrust has been a hot topic — where do you see governments drawing the line between protecting consumers’ data and allowing companies to add value for consumers in innovative ways? And where do you, personally, draw the line at Price.com?
RJ: In the context of Silicon Valley, there has been a lot of concern that tech giants are gathering too much personal user data and potentially selling it. The government is drawing the line by implementing stricter regulations to promote transparency between consumers and businesses when forcing laws like GDPR, so users are fully informed on how the data is being utilised.
Protecting user data and showing consumer privacy is a top priority and core value at Price.com, and while we improve our technology and platform by aggregating data across all consumers, we will never sell personalised user data.
RJ, Marco, what do you think are the major threats to AI companies in terms of IP and disclosure? Or do you believe that the deep technology, itself, is a large enough moat?
RJ: When filing a patent for AI technology, much of the process has to be disclosed, from how data is collected to how it's processed. However, disclosing the process also allows others to replicate it.
I heard this story that Coca Cola never patented their secret formula to ensure that competing beverage companies couldn't replicate the recipe.
I believe that filing patterns to protect IP is an important part of companies’ development and growth, however, it takes an immense amount of time, money, engineering talent and effort to grow an AI-based company from the ground-up and is not easily replicated — with or without production of the core technology or the platform. So I feel like deep technology serves as a large enough moat to justify that.
Marco: It's always a danger to have something so advanced and in my experience, customers and their engineering teams would want to know how the technology works. CHRONOS has been granted 13 patents for our algorithm, and even if someone were to steal it, it would take many years to reengineer or surpass our capabilities. I think we feel very safe today opening up what we have to customers, and they would rather work with us than trying to copy us.
Terry, how do you think about IP protection when looking at AI companies?
Terry: Even if software companies were to patent their technology, they wouldn't patent the entire lifecycle, especially the secret sauce of their technology. This enables them to protect the most proprietary part of their base codes, and sometimes even restrict access to it within their own organisation as an added layer of protection.
Hardware products are probably more prone to patent and IP protection, simply because hardware technologies are more tangible and easier to patent and protect, versus for software which is more intangible. Even if you submitted a paper that has your methodology, it doesn't mean your IP is tangentially protected under the patent.
And I think your patent strategy is almost determined by the location that you are in. We like to see our companies’ technology patented as far as possible.
But there are also companies that are in China where your trade secret is way more valuable than having to publicise and patent your technology. There’s an interesting example in our portfolio of a company that provides state-of-the-art picking solutions for e-commerce using robotic software. They are already deployed in the US and Japan with their partners, but they would be very cautious when entering the Chinese market, especially if they were asked to deploy on premises, because there's also the risk of reverse engineering. And although it might take them years, it will still make their use case obsolete — and that is the danger right there.
In those cases we would support their decision to deploy in markets that they feel safe in with good partners, prior to tackling China, which is potentially the biggest e-commerce logistics market that they could enter.
The point is — choose your partners and choose your playgrounds carefully.
What is your take on the growth of AI in general in Asia, especially since countries like China, Japan, South Korea and Singapore have all been making major strides in the AI space?
Marie: While I can’t speak for all countries, I'm certain that Singapore has a high potential in AI. There is a big push from the government and large inflow of investment (public and private) in Singapore. There is also a structure and focus on innovation with well-supported incubators. Other countries like Vietnam and the Philippines have also developed their IT resources and skills and are one step away to moving closer to AI-readiness.
RJ: We are looking to enter the Asian market. Asia is now online both on mobile and desktop, and we are seeing emerging web-based technologies experience rapid adoption rates throughout Asia. Our primary goal is to solve problems. We already have a large portion of engineers within the company based in Asia right now, and with so many companies transitioning into being remote-based, it has become far more standard to maintain a global team, and I feel like some of the best and brightest people are based in Asia right now.
Marco: I know Singapore well enough to say that I'm very impressed with the efforts made to be at the edge of technology while trying to create and invite creativity.
Terry: Regardless of the downsides of the lack of IP protection in China, I believe China remains the largest AI player in the world, and there are many AI startups that are positioned to capture that within them. One of our portfolio companies Sensetime has 3,000 patents and are ready to operate in the market just based on just their application strength.
China is very advanced in applications — even more advanced in the US even though the US has better talent, research methods and capabilities in writing better algorithms. I think what has happened is that there has been a push from the government to combine public data with AI technologies to enable the algorithm to learn faster than any other country. For example, Chinese facial recognition is able to beat Facebook's visual recognition in terms of accuracy because of the gigantic volume of data they are able to consume and analyse versus in the US, where all facial recognition data would have to be privately sourced and Facebook would be the largest network with ready accessibility to that.
I think China will remain uncontested as the largest AI giant in the world, with BAT — Baidu, Alibaba and Tencent — driving the AI Innovation wave. I don't see the same in other parts of the Asia today.
Let's talk about countries that may not have the same sensibility as China — like Japan and Korea. I think from an application perspective they’re nowhere near China's pace of development, but I think they are well positioned to capture industrial automation because of the nature of their countries — they have an ageing population, limited workers and a lot of automation capability and are hence ready to capture any industrial AI in innovation.
Terry, could you elaborate on the application of using AI in the area of cybersecurity?
There are many aspects of cybersecurity that make use of AI. For example, companies like Symantec and McAfee never really used AI — they have a manual library which is updated by their engineers to match zero day attacks. One of our companies, Deep Instinct uses deep learning to automate that process and can consistently beat the accuracy of the likes of Symantec and McAfee.
There is also anomaly detection — an AI-based behavioural analytics method to capture how hackers perform or what areas they are trying to attack within your system to understand and eventually automate that.
Marie: Edge computing is also one of the aspects where AI is also very much helping. It is a domain where a lot of research is being done and while there are a lot of improvements still to
Terry: Fully homomorphic encryption, which is a leading area in encryption I think will change how data works in the future and it will make consumer data much safer.