Harnessing Digital Innovation for the Growth of the French Economy
Digital innovation largely depends on the development of artificial intelligence. It is therefore essential that research, education, companies and the government in France adapt to this new reality in order to strengthen the country’s economy and allow it to stand up to American competition in this domain. Eric Chaney, economic advisor to Institut Montaigne, identified for the “Rendez-vous de Bercy”, an event hosted in November 2017 by the French Ministry of Economy and Finance, ways for France to become leader in artificial intelligence, and thus in innovation.
1. Innovation has numerous facets and Artificial intelligence (AI) is essential to most
The current stream of scientific and technology innovation is very broad – high throughput material science, single-cell genomics, boosted by cheap DNA editing techniques, disease detection, new drugs, computer science, behavioural economics and finance, marketing, corporate strategy … Yet, more and more fields are using Artificial Intelligence (AI), and more specifically deep machine learning (DML) as a tool to leverage and accelerate R&D and, as far as business is concerned, to gain market shares and market power while cutting costs. Therefore, it makes sense to focus on AI when addressing today’s digital innovation.
2. Taming US giants may be desirable, but it won’t change the competitive landscape
Giant technology companies such as Google, Facebook or Amazon have already invested huge financial and human resources in DML developments, as exemplified by Google’s acquisition of DeepMind in 2014. They will continue to do so in a foreseeable future. Should Europe aim to foster its own giant tech companies, including if this involves challenging the business of these US companies, which have become quasi monopolies in some sectors?
EU competition authorities are moving in this direction by cracking down on these companies’ tax liabilities or by imposing tougher rules regarding private data protection. Yet, while it certainly makes sense to enforce EU and national tax as well as competition laws for all companies, however shrewd they may be in the art of tax-optimisation, this is unlikely to significantly change the competitive landscape. Indeed, the fuel that feeds most ML algorithms is data itself, which tech giants have plenty of and are accumulating faster than any competitor thanks to their market share.
Ring-fencing the access to European consumers and companies, as China does, thus allowing its AI-intensive enterprises (Baidu, Tencent) to thrive, while restricting their ability to penetrate Western markets, is obviously not an option for an open European Union. Fortunately, the future is not locked-in: as History has shown repeatedly, the scope for the next disruptive innovations is wide open. In addition, markets may be more contestable than they appear at first sight, as OECD researchers have demonstrated in a recent study: “only marginally superior products can take over the entire market, rendering market shares unstable”. Therefore, Europeans should ask themselves: what should we do to foster indigenous, AI-intensive, digital innovation?
3. Artificial intelligence: user or leader? It all starts at school
At this stage, only narrow AI (which covers a limited field of knowledge, no matter how immense and complex it may be, is available or foreseen. General AI is still many years, perhaps eons, away (4). Yet, the scope of narrow AI is already very comprehensive and will expand further, as DML continues to be perfected. AI-enhanced applications (apps) have begun to reach the market and will continue to spread in the coming years. Billions of potential users are, knowingly or not, waiting for them. The question is: should we, Europeans, resign ourselves to be users only, perhaps followers (duplicating innovation by adapting it to local features), or should we rather aim to join the leading squad?
Since it would be reckless to predict, today, what exactly our students should learn to later become leaders in this field, the optimal strategy must include enhancing scientific education – maths, physics and computer science — for all students, from high school to college. Note that a solid scientific education requires a good grasp of common language, reasoning and history, which provide a natural bridge across the often-feared gap between humanities and science. Initial education, but also vocational training, are both at stake. For instance, engineers of previous generations need constant retraining. One perhaps extreme view is that only the basic science cognoscenti will be able to grasp AI developments smartly enough to keep their head above water when our lives and businesses will be flooded with AI apps. STEM education, which, as shown by the steady decline of French PISA scores, is a serious issue for France, makes the country’s response to the AI challenge even more pressing. This is a problem which we should tackle at its roots (see Institut Montaigne report for an articulated digital strategy starting in primary education). Here is a good starting point: understanding that maths and science are not only made of logic and abstract concepts, and trying to see them as they are: beautiful and exhilarating, as Nobel Prize winner Georges Charpak used to say.
4. Companies are at the forefront of the AI battle, because competitiveness is at stake
In the corporate world, AI development is not only reserved to the happy few, such as high tech companies, or to top engineers, CIOs, CTOs, CFOs or other C-level managers. Harnessing AI will be a matter of competitiveness before anything else. Companies smart enough to provide the whole range of their staff and, in some cases, their clients with AI-enhanced apps will gain a competitive advantage and eventually outsmart competition. From top to bottom, from assistant level to back offices to front offices, HR departments and top management, AI will make the difference between winners and losers, because early adopters will increase their productivity by raising the value generated by each employee, while laggards won’t. Winners will gain pricing power and will be able to hire more and/or raise their employees’ compensation, thus attracting high talented workers, while laggards will have no other option than to cut costs by laying off employees, and will eventually end up priced out.
From this angle, the innovation battle is pretty much a bottom-up story, in which governments seem to have little to say. Is this really the case?
5. How governments can and should help: competition policy, balanced data protection policy, public data policy and public procurement
- Keeping the playing field levelled
First, innovation is often stifled by the market structure. Indeed, large companies already benefiting from a large market share, such as incumbents in the tech area for instance, have no interest in letting innovators reach the market. If they do, it would probably be more out of an interest to buy them out than to surpass them. This is well known in the field of patents, where patent trolls are an extreme, yet worrying example. This is a window of opportunity for governments to intervene and keep a level playing field open to innovators, even if it is at the expense of well-established names, which are always quick to threaten to cut jobs if their turf runs the risk of being invaded. Well-designed competition and patent policies are needed in these cases, both at the national and European level.
- Balancing data protection and data access to researchers and companies
Second, we are lucky enough to be at the very beginning of the AI era. At this stage, AI algorithms need to be tested on large data sets before they can beat human experts. Take medical applications for instance. Europe, with its highly-sophisticated healthcare systems, and its detailed patient records, is sitting on a gold mine. Of course, one might wonder how to make these data available for researchers as well as private companies developing disease detection algorithms, without compromising data privacy. This difficult question must be addressed heads-on. Since the protection of privacy is a highly sensitive topic for European citizens, perhaps more than for their American peers, their trust is a necessary condition to the flourishing of AI-based innovation in Europe. Strong and transparent EU and national regulations are therefore needed. At the same time, excessively convoluted protection and, worse, inconsistencies across countries, are likely to hamper AI-intensive innovation, at the expense of Europeans’ well-being.
- Making good use of government data banks
Third, government themselves own very large datasets, through their own constituencies. Administrative records, be they from hospitals, or other types of national administration, have most often already been digitized, or are in the process of being so. Since large and accurate datasets are critical to feed AI algorithms, governments should think of allowing researchers and private companies to access them, under a license or for free (one may argue that the social return on AI investment on disease detection is high enough for the latter option), provided that individual data protection is warranted. Scientists are used to working on anonymized datasets for double-blind experimental procedures. There is no reason why safe anonymization could not be used for AI-based innovation.
- Using public procurement to foster AI-based innovation
Fourth, governments have a powerful tool to encourage and foster AI-based innovation: public procurement. Here are a couple of ideas where open calls for tenders could kick start new industries:
- Day-to-day government services. Most government services are labour- rather than capital-intensive. A significant part of the paperwork performed by government employees could either be done by AI-enhanced softwares, or at least boosted by such algorithms. As previously mentioned for the corporate sector, all employees, whatever their training or grade, are eligible to benefit from innovative softwares designed to improve the quality of their work and to allow them to focus on tasks they excel in.
In this regard, the right question to ask is not: “What can humans do that intelligent algorithms can’t”, because the answer will always be contingent on the state of the art. It should rather be: “What can intelligent algorithms do that humans do routinely?”
Why not ask secretaries and assistants to participate in the selection juries? Once again, boosting productivity should not be restricted to top experts and managers only. It is worth noting that such measures could allow governments to hit two birds with one stone by both reducing civil service headcounts and making government employees not only more productive, but also prouder of their job.
- Medical monitoring of the population. Most European countries, including France, benefit from highly-sophisticated networks of public and private hospitals, and of independent medical doctors and nurses. Early detection of cancers and retinal degeneration via AI algorithms is already a reality. Based on public procurement competition, the government may encourage research and innovation by making the use of such apps mandatory, once they have been validated by scientific authorities. This would also save future taxpayers important amounts of money, as it would allow for a more efficient prevention process and for early disease treatments.
- Education. Correcting their students’ essays remains one of teachers’ most important and yet most time-consuming and boring tasks. AI algorithms could considerably help them to reduce time wasted in proofreading and allow them to focus on what matters most instead: the value of their students’ essays and what they say of the students’ learning curves. In some cases, MCQs for instance, pieces of software could even oversee the correction of examinations.
- Public security. A growing share of public spaces is monitored by CCTVs, be they owned by local governments, transportation companies, or private companies. Face recognition apps, an area where deep learning has produced remarkable (although not fault-proof) results, could be used to spot dangerous individuals, thus helping law enforcers to prevent crimes and terrorist attacks.
- ‘The new frontier of genome editing with CRISPR-Cas9’, Jennifer A. Doudna, Science vol. 346, 28 November 2014.
- ‘Digital Innovation and the Distribution of Income’, Dominique Guellec and Caroline Paunov, NBER Working Paper No. 23987, November 2017
- ‘What is Consciousness, and could machines have it?’ Stanislas Dehaene, Hakwan Lau, Sid Kouider, Science vol. 358, 27 Octobre 2017
- ‘Le numérique pour réussir dès l’école primaire’, Rapport de l’Institut Montaigne, mars 2016.
- ‘Patent Assertion and US Innovation’, Executive Office of the President (President Obama archives)– June 2013
- ‘Discrimination of Breast Cancer with Microcalcifications on Mammography by Deep Learning’, Jinhua Wang and alii, Nature Scientific Reports, 6, 27327, 7 June 2016.
- ‘Deep learning approach for diabetic retinopathy screening’, E. Colas, A. Besse, A. Orgogozo, B. Schmauch, N. Meric, E. Besse, Acta Ophtalmologica, Vol.94, Issue S256, October 2016.