The AI trilemma: Saving the planet without ruining our jobs
Digitalisation and artificial intelligence affect more and more areas of the world of work. Rising risk of massive job losses have sparked technological fears. Limited income gains and productivity gains concentrated among a few tech companies are fuelling inequalities. In addition, the increasing ecological footprint of digital technologies has become the focus of much discussion. How can this trilemma be resolved? Which digital applications should be promoted specifically? And what should policymakers do to address this trilemma? This contribution shows that policymakers should create suitable conditions to fully exploit the potential in the area of network applications (transport, information exchange, supply, provisioning) in order to reap maximum societal benefits that can be widely shared.
The rapidly growing number of artificial intelligence (AI) applications in the world of work has renewed fears of technological unemployment. Whether autonomous taxis, fully automated logistics centers, the Robo-Hotel concierge Pepper or the Bar Tender Tipsy Robot; in more and more areas machines seem to be able to replace us. This is especially true in those areas where we ourselves have been convinced of being irreplaceable: In artistic or intellectual activities.
As AI enters our lives more and more, however, the original fear that this would render human abilities superfluous has given way to the insight that it can be used in a wide variety of ways. In addition to replacing work, digital applications can also lead to job enrichment, which should tend to allow workers to command higher incomes and enjoy higher productivity. Most importantly, the use of AI will gain significant in areas where no human or machine could be used efficiently in the past: In the control of networks, for example in transport, in waste management or in information exchange.
Modern urban traffic control systems can use flexible traffic management to direct individual and public transport in such a way that the traffic volume is managed optimally and efficiently. AI will also become increasingly important in the area of electricity network control, especially where more and different energy sources have to be connected when economies are transiting to sustainable energy supply.
Even in the world of work, such AI applications are becoming more and more pervasive. For example, computer routines can pre-select job applications, carry out selection interviews in a semi-autonomous manner and, after successful recruitment, also automatically produce personnel evaluations based on a variety of indicators.
The AI trilemma
What all applications have in common, however, is the enormous consumption of energy. From the storage of data in cloud computing centers, to data analysis by high-performance computers, to the power consumption of even the smallest mobile digital devices needed to stay connected, the digital economy is already using up more than 6 percent of average electricity consumption. And the trend is rising sharply. Without major efficiency gains, electricity consumption is expected to rise to over 20 percent in 2030.
One area where this has already led to serious problems concerns the blockchain technology, and particularly its applications in the field of cryptocurrencies. Bitcoin, the most well-known crypto money, and its underlying protocol currently requires the power consumption of a country like Denmark. This is unlikely to be a sustainable concept.
There is another problem: Many digital processes require data and analysis to be concentrated at individual developers and companies. In addition to the enormous productivity gains that individual companies can achieve — see Amazon.com — this also leads to a further increase in global inequalities. Increasing market concentration of digital companies and widening income differentials then prevent stronger growth for all. Digital growth is not inclusive and — depending on the application — it is not resource efficient.
At the micro level, too, problems are emerging that perpetuate existing inequalities. In order to train AI routines, historical data must be used, which often reflect discrimination, such as a disadvantage of women or ethnic minorities in the labour market. If an AI-routine is fed with such data without a corresponding filter, the disadvantages will be continued in the next personnel selection. Amazon has already experienced this to its disadvantage.
This productivity paradox — low growth, greater inequality and high energy consumption despite rapid technological progress — is mainly due to the digital character of AI: silicon-based information processing consumes resources that can only be optimized and reduced to a certain degree. And poorly enforceable property rights in the digital economy determine corporate strategies to hoard data for better exploitation without further access. This “weightless economy” now occupies the largest place and leads to market distortions that have so far received insufficient attention.
How can network application address this issue?
Nevertheless, not all AI applications are affected to the same extent by this trilemma. Especially the above-mentioned network applications have the potential to perform particularly well when it comes to low net resource consumption and high inclusivity. Well-trained AI routines, for example regarding electricity management or water consumption in agriculture already reduce the burden on the environment today and offer possibilities to address climate change effectively. Furthermore, such solutions also offer opportunities for cost-effective knowledge transfer to developing countries, where there is still a great need to catch up on modern technologies adapted to local conditions. Companies such as Google and Microsoft have already discovered this need and have begun to establish their own research centres in some developing countries. And local solutions, especially in agriculture, also show significant potential for more productivity gains.
Nevertheless, the AI trilemma is still not resolved and the risk of deepening inequalities and job losses remains. The latter is particularly important in developing countries, where competitive advantages have so far consisted of simple activities that can be automated quickly. Already during the earlier wave of robotization, most jobs were lost in these countries.
Moreover, the network character of these applications leads to further concentration and market power. Providers who bring together the largest possible volumes of data from different sources have the advantage of offering cost-effective solutions and thus pushing smaller, specialized companies out of the market. The inequality problem of the AI trilemma thus remains.
What can policy makers do?
In the current debate, three main approaches are being discussed to resolve the inequalities of the AI trilemma:
A first, traditional answer is to try to use taxes to better capture capital gains, while at the same time shifting the tax pressure from labour back towards capital. This has often been discussed in connection with a robot tax. On the one hand, it would allow the enormous profits of digital companies to be captured. On the other hand, tax fairness would be restored, which could relieve the factor labour and ease the pressure towards rationalisation and job losses. However, in a global economy, governments have tight limits on how much they can tax internationally operating companies. Attempts to extend taxation to the consumption of digital services instead of profits are being resisted by those countries that are home to a myriad of large, digital companies.
A second, more innovative approach is to ensure greater competition between digital enterprises by making it easy to transfer data between platforms using uniform standards and protocols. Some solutions also propose data ownership in order to provide a monetary incentive for those who make their data available by using the platforms. So far, however, these solutions are not yet fully developed and practicable. Moreover, only very few users can derive relatively large profits from such approaches, while the vast majority of them would have little to expect. The incentive to switch platforms or to reap monetary rewards would be too low to solve the AI trilemma.
A final, little debated solution is to set up a sovereign digital wealth fund that participates widely in the digital economy. Such an approach already allows the benefits of public goods — in this case freely available information — to be passed on to a broad group of people. However, instead of feeding off oil wells (Saudi Arabia, Norway) or fish stocks (Alaska), such a digital wealth fund would be financed by taxes and new debt, in order to generate profits by investing in a broad fund of innovative digital companies. At the same time, such a fund, provided it invests deeply enough, would also be able to directly influence the operative business in market-dominant companies in order to prevent the exploitation of such positions. At the same time, the fund could also aim at exerting influence at the micro level to ensure that ethical and ecological standards are met when using AI.
None of the solutions outlined here will be sufficient in themselves to resolve the AI trilemma. National solutions often do not provide sufficient guarantee that all market participants will actually be offered the same conditions. International approaches, especially in the area of taxation, are slowly gaining acceptance, but often only at the lowest common denominator. Innovative solutions such as data ownership require institutional changes, which will most likely take some time to become established. However, an approach that addresses all three proposed solutions should make it possible to find initial answers to the AI trilemma while at the same time offering new, individualised proposals that optimise the potential that AI holds for jobs, income and inclusiveness. The future of work demands not only technological innovations, but also political and institutional ones.
