Many fields of artificial intelligence in all areas of human life are experiencing a tremendous experimental boost during the corona crisis. These experiences will significantly accelerate the subsequent development. The leaders in Germany and Europe are now being challenged.
When SARS broke out in 2002, it took months to sequence the genome of the virus. The coronavirus that causes Covid-19, on the other hand, was isolated and sequenced with the help of artificial intelligence (AI) by a team under the direction of Professor Zhang Yongzhen at the Shanghai Public Health Clinical Center on January 5, 2020 - two days before China even officially announced the existence of a new disease. On January 11, the team published the results and made the data available to the global research community.
AI is also playing an essential role in many other activities in the management of the corona crisis in both the business and health sectors. This has made two things clear:
No one can overlook the enormous power and necessity of AI, and China's lead in some fields of application immediately became apparent - perhaps for the first time for many. The coronavirus is an experimental super test for AI: Many of AI's large-scale applications would have been unthinkable on this scale without the coronavirus. Other AI applications already in widespread use today are going through a kind of “black swan” test with the virus. Users are learning a lot from AI performance in both cases. The above-mentioned developments have the potential to tremendously accelerate the development of AI in the wake of the coronavirus.
AI's fields of experience in the corona crisis can roughly be divided into three areas:
Crucial analytical acceleration and scaling
Fields of application that can no longer be attended to by humans
Applications for which the extreme events present a dynamic “black swan” test.
All three result in obvious fields of action for companies and government agencies, especially in Germany and Europe.
Analytical acceleration and scaling
China has invested enormously in AI in recent years and is now the world leader in AI application. A few weeks ago, the Chinese technology company Baidu described in MIT Technology Review how it applied AI to help fight the coronavirus:
Use of its LinearFold algorithm to determine the secondary structure of viral RNA in 27 seconds;
infrared sensors that perform multi-person temperature control without contact and are used to measure a fever at the Beijing train station;
the first open-source product that can declare with 97.27% confidence whether all individuals in a crowd are wearing oronasal masks properly;
diagnosing Covid-19 through breast CT lesions by means of LinkingMed, which uses Baidu's deep-learning platform PaddlePaddle, with 92% accuracy (and recall of 97%) at XiangNan University Hospital;
tracking people's movements out of Wuhan via Baidu Maps and more ...
Of course, Baidu is not alone in this. Insurance giant Ping An runs the most successful AI-assisted telemedicine platform, Good Doctor - with 300 million users at the end of 2019. It handled 1.11 billion requests at the end of January/beginning of February. While Good Doctor previously focused predominantly on 24/7 issuance of repeat prescriptions with 1hr delivery times, regulatory restrictions were lifted during the crisis permitting AI-assisted initial diagnoses in teleoperation.
Additionally, Alibaba has not only sent millions of masks and test kits around the world but, similar to Baidu, has applied AI-based methods in research, test development (at its Damo Academy) and patient counseling. Tencent and many smaller Chinese companies, such as Infervision, have taken similar approaches to applying AI to aid during the crisis.
The leading AI companies from the USA are also reporting on their endeavors with motion analysis and medical aids, albeit with an incomparably weaker impact than the Chinese companies thus far. To date, no major AI-based initiative from Europe has been recognized in the management of the corona crisis.
The jackpot, of course, is the development of a corona vaccine or at least an effective treatment. Not only are the top medical laboratories and tech companies working on this, but many AI startups as well. These are more globally distributed. For instance, IEEE Spectrum references Dargen (South Korea), Insilico Medicine (Hong Kong), SRI BioSciences (Menlo Park, USA), Iktos (Paris) and BenevolentAI (UK). They use AI-based screening methods to identify the most promising among known candidates and move them to the testing phase. Or they are working hard to find new active ingredients using cutting-edge AI.
Fields of action
European health sector: Many of the AI-assisted actions implemented in China would have helped greatly in Italy, Spain and other regions as well. Even if Germany hopes to come away with minimal lasting repercussions due to advance warnings, government agencies should ask themselves whether we should not gain speed, scalability and cost efficiency through a more innovative use of AI - all of which are crucial goals even beyond the pandemic.
European Commission: So far, the concrete results of the European AI initiative are largely limited to ethical principles and regulations - the supposed “protection of citizens” takes precedence over innovation. With all due respect to the work of the Commission, questions of balance must be permitted here:
Can we afford not to adopt AI tools? Where is the collective well-being more important than individual fears?
Doesn't this mean that we, here in Germany and Europe, are on the way to becoming an AI colony - and will thus lose any opportunity to shape development?
Is there not a need for decisive support for the multipliers of applied AI, complemented by government mandates for innovative AI startups and access to mass data?
Companies: Every single company should take the above-mentioned experiences to heart and check whether the company itself has already decisively gained speed, effectiveness and scope through AI applications in offers and processes. Otherwise, it is a matter of upgrading quickly.
Machines instead of people
Fields of application such as telemedicine not only have the advantage of improved scaling but also of managing their tasks without physical contact - invaluable in the corona crisis. The Italian doctor who infected hundreds of patients comes to mind, or the French and Chinese doctors who died as a result of Covid-19 transmission from patients. As reported by the South China Morning Post, the first autonomous systems and robots were also used in the country to disinfect isolation wards and deliver food and medicine. Even the sometimes controversial nursing robots in Japan are certainly enjoying unaccustomed popularity these days.
In other areas, people could no longer handle the sheer volume. For example, Baidu also used a robocalling platform to ask people to volunteer travel and contact information in over 3 million calls (1,500 calls per second).
Even in companies only indirectly affected by the crisis, AI must step in: Facebook famously became one of the first companies to respond to Covid-19 in late February, allowing its 45,000 employees work from home. All employees? Well, 15,000 contractors had to stay in buildings all over the world: They identify and remove “objectionable content” (primarily sex- and violence-related but also “coronavirus fake news”). Since these people work with highly sensitive user data, they need to be behind the firewall. Only after 4 more weeks, on March 24, this group was also sent home in the face of the surging crisis - still paid, although they could no longer work. The bulk of the work must now be performed by AI. Although it is known that these AI systems are still occasionally overwhelmed, the volume can no longer be handled any other way. Apparently, a few AI errors are still better than complete passivity. It will be an unplanned giant stress test for the algorithms - even at Google and other companies struggling with similar difficulties.
There are many other examples of how digital, AI-driven services are increasing and being put to the test during the crisis. Perhaps somewhat unexpectedly, the use of AI-based recruiting tools has skyrocketed: Curious Thing, Talview, Knckri, Eightfold, Ziprecruiter and more report soaring sales. The coronavirus isn't exactly currently being considered a source of a new miracle in the job market. On the other hand, a comprehensive hiring freeze is not affecting the market - Amazon alone, for example, is hiring 100,000 employees (temporarily) and welcomes people from the hard-hit tourism industry in particular. However, no company is currently risking extensive in-person interview processes, and it is precisely in the initial screening and digital appointment setting process that machine intelligence is welcome.
Fields of action
Companies: Even without the corona crisis, there are many fields of work that are very dangerous for people (not only in mining and oil & gas), that lack skilled workers (such as nursing) or that cannot be performed by people at all or only at low wages due to extremely tight margins (such as call centers). This often results in output loss or long waiting times. AI-based systems could be used in many situations - at least as a back-up option - and soon these systems may be able to completely take over some tasks.
Corona crisis as a black swan stress test.
Additionally, experimental data is currently being collected with the use of AI in a completely different field. In many places in this world, AI systems work semi-autonomously and try to adapt dynamically to new developments. Their robustness of several is being fundamentally tested by the coronavirus: Probably no one, at least outside of the WHO and isolated crisis teams, has trained for a scenario such as this in advance. One must observe how such AI systems react to the black swan events. Presumably, this is where the wheat will be separated from the chaff in terms of AI maturity.
What does one do as a user of an AI system when such unforeseen events occur? Often, the reflex, as in an airplane emergency, is to turn off the autopilot and take over manually. However, as with pilots, few have an emergency checklist at hand or have trained extensively in its use. In the worst case scenario, one retreats from data-based decisions and cedes to the gut decisions of HIPPOs (HIghest Paid Person in the Office) - hardly a recipe for success. More well-prepared companies have a tool and process landscape to support humans (as in “human-in-the-loop”) to intervene effectively and quickly in AI.
In some environments, that is not going to be easy. Financial markets are such an example. Transactions can be switched to manual operation, but only at the cost of several orders of magnitude in speed and complexity mastery - which can be tantamount to a death sentence in a highly fluctuating market. There will be plenty of opportunities for analysis and learning in this after the crisis, particularly since individual investment banks and funds seem to be bucking the trend and performing excellent trading.
Currently, many companies also tend to place little trust in AI-based market forecasts. AI can now be built in very different ways - there is even high intelligence and low intelligence. Modern “adversarial” methods often employ machine adversaries which use artificial intelligence on their part to strengthen the system against extreme events beyond the scope of the training data before the real emergency occurs. The minimum requirement should be that systems shall be fully operational again when economic activity resumes and that they were not designed so statically that they need to be trained from scratch. Again, in the rear-view mirror of the crisis, the AI degree of maturity within companies becomes visible.
Interestingly, DARPA, the US agency that has given the decisive impetus to many pivotal innovations that have inspired more progress, already announced a new competition last year: SAIL-ON (Science of Artificial Intelligence and Learning for Open-world Novelty). This competition aims to promote research and development of scientific bases, engineering techniques and algorithms for AI systems to enable them to cope with highly dynamic situations. The intended purpose here was the use of AI in war situations. However, it would not be the first time that the essential contribution lies in its application in civil society, which is hardly less dynamic today.
Fields of action
Companies must learn to build or use robust state-of-the-art AI systems and recognize their strengths and weaknesses. Effective tools should then be used to manage the weaknesses - even if it is to support the “human-in-the-loop.”
Germany should ramp up its own Federal Agency for Disruptive Innovation considerably faster and more autonomously, and AI issues need to become a central focus.
The path forward
The corona crisis highlights how the professional application of AI is already leading to critical systemic advantages - with China as a prime example. Nevertheless, these are early phase developments with shortcomings - comparable to the use of the German Air Force, or Luftwaffe, in the First World War.
The ensuing turning point and triumph of aviation may now be upon us in the use of AI as well. Chinese and US tech players have had such tangible first hand experiences that they will surely redouble their efforts. We expect that even beyond this, the adoption rate and quality of AI applications in a post-corona world will definitely skyrocket based on these experiences.
Germany and Europe are certainly not in the front row at the moment. Nevertheless, every crisis is also an opportunity: Precisely the standstill of many routine jobs in this country can be used for wide-range AI training of employees as well as for re-planning a future AI strategy. This is not only about process optimization but also about new business models and often decisive competitive advantages. In the post-corona world, many cards are being redealt. The hope is that, from healthcare to industrial applications, the lessons of the crisis will yield a drive forward a widespread and intelligent use of AI in this country as well.
Dr. Philipp Gerbert
Dr. Philipp Hartmann
Director of AI Strategy
Dr. Andreas Liebl
Managing Director ; Founder