On December 31st 2019, a small Canadian Artificial Intelligence startup called BlueDot, which uses natural language processing and machine learning to scour through hundreds of foreign sources daily, issued a thorough warning that an “unusual pneumonia case”, originating from the Wuhan province in China, would be spreading across the globe in record time, and it projected the routes it would take to do so. And when it does, it would be a killer.
This was more than one week before even the (until recently…) well-funded, very powerful World Health Organization (WHO) knew much about it. The WHO announced a “flu-like outbreak in China” on January 9th, a full 9 days after AI company BlueDot knew about it.
Yeah, we all know about that. And?
This was a great first start: Artificial Intelligence was beating older systems of pandemic detection, and things were promising. After all, we’ve been spouting off about how much progress we’ve made in AI research and development.
Surely, AI would come to the rescue to combat what we now know and fear as COVID-19, right?
Before you and I were politely asked by our governments to please stay the hell at home and put our businesses and lives on hold, there was a time when the novel coronavirus was nothing more than another medical nuisance from China.
Fast forward a few months and virtually the entire world has changed, and will likely continue to feel its effects for generations to come.
If ever there was a need for Artificial Intelligence to deliver on its promises to save humankind, this is it.
But besides BlueDot’s incredibly accurate prediction, what exactly, are AI researchers and scientists doing to combat this invisible menace, and help bring us back to normal?
If you love the promise Artificial Intelligence holds — like I do — and understand its potential, you won’t like the answer, if you haven’t gleaned it for yourself already. Below follows a highly oversimplified, not-at-all comprehensive, probably already outdated rundown of some of the highlights of AI vs COVID-19. Outdated because in AI, we are looking at moving 2 steps forward, 1 step back, every day. And sometimes it’s 2 steps back.
Diagnosis – it’s where it all starts
The world was woefully ill prepared for this virus. Once we knew about it, we didn’t know how to test for it. Once we (thought) we know how to test for it, we didn’t know how to make enough reliable testing kits. Once we (thought) we know how to make them, we didn’t know how to distribute them and train healthcare professionals in using them.
Viruses are both very simple and very complex. Knowledge on how viruses “work” from previous research has helped a lot here, but every virus is different. The thing we’re fighting as of the time of this writing is called the “novel” coronavirus for a reason — ”novel” , in this sense” means “new”.
AI to the rescue — let’s have the machines figure this thing, out, right?
How did we utilize Artificial Intelligence in figuring out how’s sick in the first place?
To my knowledge, Chinese company Infervision was the first to launch a medical imaging platform based on deep learning that accelerates the detection of damage of the lungs potentially due to the virus. Infervision began using their system at the epicenter of the outbreak, Tongji Hospital in Wuhan, China.
Since then, doctors and scientists at Wuhan’s People’s Hospital, under the leadership of Dr Lin Li, created a neural network that extracts visual features from chest CT scans and can differentiate the ones of a coronavirus patient from one affected by “regular”, good ol’ pneumonia.
Alibaba, often referred to as the Amazon of China, came up with a similar detection system using AI. It can perform a test in about 20 seconds instead of the 15 or so minutes it would take a human to diagnose patients. Accuracy? About 96-percent, which is on par or higher than a human doctor’s ability.
And then there is Chinese Internet giant Baidu which created an AI system using cameras equipped with infrared sensors and facial recognition. Baidu’s system can detect people’s temperatures in public places such as airports and bus stations, around 200 scans per minute. In China’s autocratic system, anyone who gets flagged as having a higher than normal temperature gets taken aside by authorities for further checking. Creepy but apparently effective.
That’s (at least) 4 wins for China, if you’re keeping score. There is more but you get the point: China was leaning in to this COVID-19 thing with AI right away.
As with everything coming out of China, however, we need to be careful about taking such information with a grain of salt. Has either technology been shared and vetted by non-Chinese entities? Some of it actually has, to my knowledge. Some has not. Whenever our office or one of our associates is dealing with anything China related, there’s a good chance information is either outdated, inaccurate, or simply false.
China works hard on winning the race to AI Supremacy. Sometimes their progress is truly remarkable. Other times, it’s propaganda.
As of May 9th 2020, the United States (and the rest of world) still does not have one uniform standard of detecting and diagnosing large numbers of the population. Sometimes doctors will do a nasal swab, sometimes draw blood . It’s still pretty much a case by case thing, and test results can take days to receive. This is not only emotionally stressful by itself, it’s made worse because in many cases, the potential COVID19 host is quarantined.
OK, what else?
How else has Artificial Intelligence been helpful in improving the diagnostic part, besides the aforementioned examples from China?
It hasn’t, really.
Having read on so far to this point, you may already sense that this is not going to be the usual AI-worshipping article you might find grinning at you from your screens, convincing you of the infallible awesomeness of AI.
No, I’m disappointed, embarrassed, and angry. We could have done better, and we should have.
AI should and could now be at the point where we can reliable “fix” things like this. But we don’t, and we are nowhere near to the point where this will be a reality. AI has proven to be a mediocre and often unreliable ally when it comes to detecting, diagnosing, vaccinating, and combating COVID-19.
While many great efforts are being made to combat this new invisible enemy, too many AI researchers – especially new ones- are still concerned about training their algorithms to make better cat photo predictions, or score higher click-throughs on ads. We are wasting time and minds on trivial things, and the failure of the AI community to make a bigger impact on the epidemic than it as is one way we are now paying for our pettiness.
This is in no way an attack on the thousands of very clever, very dedicated folks who dilligently work on creating Artificial Intelligence that can help us through this.
This is an attack, however, on our political leaders: I posit that the lack of advancement in AI is in large part due to our leaders’ severe myopia about technology, and Artificial Intelligence in particular.
Imagine this: the U.S. Government sent $300 BILLION out in the wild to back loans to small businesses. This was a failure: the money was gone in less than 2 weeks, and only a tiny fraction of businesses that need help actually got anything. Now our politicians are gearing up to spend many more billions, likley to the same effect.
What if we had invested those $300 Billion into AI research 10 years ago? My bet is that we wouldn’t be running around with face masks, worrying about how we’re going to feed our families.
We actually did this before when we felt something was important enough to play for all the marbles: The Manhattan project (yep, the nukes) cost taxpayers the equivalent of approximately 1% of its GDP at that time. The Apollo mission (“putting a man on the moon and returning him safely to earth”…) was closer to 2.2% of 1965’s GDP, and at its peak consumed about 4.4% of the federal budget.
That’s a lot of money, but we felt it was necessary. If we want and feel we need something done, we find the means to do it. So why not with AI?
It’s clear that our current politicial leaders don’t feel that Artificial Intelligence holds promises akin to developing nuclear bomb technology or putting a human on the moon. Considering the incredible progress we’ve seen in AI already — mostly due to private sector investment, this is surprising, to put it mildly. Our dear leaders must be missing the daily reports on AI’s potential. They must be missing the countless books, TED talks, scholarly articles, and peer reviewed publications on what the future holds IF we invest the resources into developing these technologies.
Finding the Cure
Billions of dollars and over a decade of trials often need to be invested before a vaccine or medicine can be released on the market. The money part almost doesn’t matter now, but we don’t have that kind of time.
So here is where Artificial Intelligence can really make a difference – speeding up the development of a cure and a vaccine.
Unlike with the diagnostic aspect, AI has indeed been deployed in several ways to help come up with both a cure and a vaccine.
Folding@Home, a distributed computing disease research software project ran by a slew of scientists and doctors from Washington University and others, and employed by any thousands of volunteers on their home computers, is actively running simulations for COVID-19 treatments, protein structure, and more. By the way, if you have computers sitting around doing nothing, please consider donating computing time to Folding@Home. Or, download the free BOINC client to easily join other projects and set your idle systems to do some good while you’re sleeping.
DeepMind, the British company that famously created AlphaGo, the AI system that beat the world’s best Go player Lee Sedol back in ancient 2016 and 2017, has not been quiet either. Now owned by Google (like so much of AI, unfortunately), DeepMind nevertheless is putting its significan computing and human power behind the COVID-19 fight as well. Of note, a deep learning system created specifically to find new information about the stucture of proteins related to COVID-19 has performed admirably, cutting down the time from months to weeks.
The most promising advancement using AI , however, may be coming from Benevolent AI. Their treatment candidate for COVID-19, baricitnib, is entering clinical testing with Eli Lilly now.
Spotting patterns in data and make predictions based on those patterns is where AI shines. As of the time of this writing, however, only one drug which has been developed mostly by artificial intelligence is even in a clinical trial phase, an OCD drug developed by AI systems in collaboration between Britain’s Exscientia and Japans’ Sumitomo Dainippon Pharma.
Their drug creation process completed in one year what otherwise would have taken several years. THAT is the kind of progress we all want from AI, isn’t it?
But this doesn’t speed up what comes next – the clinical trial phases. In the United States, new medicines must pass 4 phases in order to get approved by the FDA. So while the speedy development process, in large part thanks to AI, is great news, this doesn’t guarantee that Exscientia or any other company can pull this off with a cure for COVID-19.
(UPDATE: Since I started writing this article, some promising vaccines and cures’ trials and approvals have been fast-tracked by the FDA)
Still, our hope and frankly our demand must be that we can use our AI smarts to come up with a cure or a vaccine or ideally both for the novel coronavirus. Several companies around the world, including Moderna Therepeutics, Atomwise, Vir Biotechnology, and Iktos are racing for the cure, and presumably to outdo the competition, generally using their own home brew neural networks to accelerate drug discovery.
Competition, in fact, may be the true motivating factor in getting this done as fast as possible. These companies are spending a lot of money, and are walking away from other potential revenue generating projects. So while I don’t doubt a general goodwill toward mankind, I suspect being the first (and only?) company to create and sell a cure is a pretty decent kick in the behind for everyone.
Massachusetts based Moderna Therapeutics might hold a secret receipe for a vaccine, mostly likely involving mRNA (look it up, promising stuff there). Secretive but clearly on to something, Moderna was slated to launch pre-clinical trials late April in Seattle, WA. With the help from the National Institute of Allergy and Infectious Diseases, Moderna might be able to speed through the testing and trial phases quicker. But will it be quick enough?
Bots and Drones
How do you get medical supplies to disease ridden areas without infecting the delivery guys and gals? Drones!
Terra Drone, among others, uses unmanned drones to ferry quarantine material and medical supplies between China’s disease control center and the People’s Hospital (sound familiar?), and they are now in other locations with large remote areas, such as Kazakhstan.
Blue Ocean Robotics has their big and cool looking UVD robots patrol hospital floors to kill bacteria and viruses using UV light (much like your fancy new air cleaner does, except on a much bigger scale).
Harvard University’s School of Public Health and Facebook, along with other partners, are working on sharing (supposedly) anonymised data about people’s whereabouts using population density maps. This, in conjunction with knowing who might have that virus, say, using thermal scanners mentioned earlier, or Verily’s temperature patch, could tell researchers (and goverments…) a lot about where people are at any given time, where they’ve been, and where they’re going.
Presumably this is helpful if you have the latter data variable – the “who’s got the virus”-part. My inner sceptic can’t shake the notion though that this tracking and sharing technology will come in very handy to just about anyone who wants to know where anyone was, is, or will be at any given time.
Yes, sorry, I’m not immune to conspiracy theories, especially my own.
Speaking of conspiracy theories, and Facebook (and Google. We always seem to talk about Google when we talk about AI…), both tech overlords are said to be working on safeguarding users from conspiracy theories (ahem!) as well as fake news and false information. Videos that offer questionable alternative cures (ie, snake oil) are supposedly removed quickly.
Who decides what’s right or wrong? In theory, vetted scientists do. In reality, though, probably some nerd working in a Facebook cubicle.
I could go on. There are indeed more examples of Artificial Intelligence being used in some form, to some degree, by different companies, governments, and individuals. The point is, did we really do all we can?
I can’t help but think “Is this it?”
OK, so AI is helping but not that much?
COVID-19 was and still is a drill for many sectors, industries, technologies, health care, and governments. Overall, the AI community has failed the test.
For the past few years we couldn’t turn around without seeing another “breakthrough” in machine learning and AI in general. But now, where it really matters, developments are surprisingly underwhelming.
Gary Marcus and Ernest Davis got it right in their book “Rebooting AI: Building Artificial Intelligence We Can Trust”. In their honest and sometimes brutal book, these AI luminaries are even more blunt than I am here. Their point essentially is that we have been approaching AI incorrectly and the types of “problems” we spend money and talent on fixing with machine learning are often either not important, could be taken care of better without AI, or are “fixed” badly.
Marcus and Davis are among (too) few reputable, high level AI researchers who admit that we are lagging behind badly and need a massive change in how we approach AI if we want to reap the benefits we’ve been promised (and promising each other) for decades.
I believe the lack of true progress in diagnosing, testing, and finding a vaccine and a cure for COVID-19 illustrates Marcus’ and Davis’ points.
In this article we discovered some companies and institutions who made some progress in the detection and diagnosing of, and coming up with a vaccine or cure for COVID-19. Some progress. This is simply not good enough in 2020, and it’s simply unacceptable going forward.
In addition to Dr Marcus and Dr Davis’ points, I feel that the U.S. and European governments are not providing enough funding, don’t open up regulations, and in short, don’t support AI in a manner that is required for true progress. China is doing a much better job at this, but I’m afraid its motives might ultimately not turn out to be so benign.
Private companies are “investing” billions of dollars into some form of AI. If you’re not doing AI, you’re out of the game, right?
The key point is that we must do AI right. We must have the same kind of funding, support, and communication the US has provided to create the atom bomb and put a man on the moon.
There are certainly a handfull of truly outstanding, remarkable companies that are responsible for the vast majority of actual, tangible progress in Artificial Intelligence. Unfortunately, those are the same companies whose profit motive has and will likely continue to drive them to only do what’s in the best interest of their bottom line, with the occasional social goodwill thrown out to the public for good PR.
I strongly believe, based on these observations, that in order for us to truly make progress in AI, we must dramatically fund non-profit, publicly accountable institutions whose sole purpose is the research and development of AI for the benefit of mankind. If this sounds like unicorns and rainbows to you, sit down for a minute and take a bird’s eye view of who currently holds the keys to the AI kingdom. Then think of which doors those keys might open. Will it be the ones that benefit everyone, or just a handful of private sector executives and shareholders?