Here - Deutscher Bundestag

Document technical information

Format pdf
Size 493.7 kB
First found May 22, 2018

Document content analysis

Category Also themed
not defined
no text concepts found


Erik Brynjolfsson
Erik Brynjolfsson

wikipedia, lookup




Deutscher Bundestag
Ausschuss Digitale Agenda
Catalogue of questions for the Committee on the Digital Agenda’s public discussion on ‘The effects
of robotics on economics, labour and society’, 22 June 2016: Professor Ryan Calo
1. The advancing development and increasing application of robotics, automation and artificial
intelligence are changing the world of work and society in general. In what fields do you see new
and/or particular opportunities and options for growth in Germany and Europe’s industries and
I am not familiar with the specifics of the German economy. My understanding, however, is that
robotics and artificial intelligence present particular opportunities in agriculture, transportation,
healthcare, and eldercare and will continue to transform the manufacturing sector. I also believe
personal and service robotics will grow into a major industry in the coming decade.
What is required, from your point of view, if Germany and Europe are to be able to stand at the
forefront of global competition in this segment of industry?
I believe Germany should continue to invest in basic research through public-private partnerships.
Robotics in particular still has far to go. For example, for drone delivery to become a reality, drones
will need better autonomous navigation, more efficient batteries, and more reliable mechanisms to
sense and avoid obstacles. Also, as robots leave the factory and enter everyday life, human-robot
interaction will become an increasingly important field.
I also believe the German government itself should accrue expertise in cyber-physical systems;
without such expertise, no government will be in a position to generate a wise law and policy
infrastructure for robotics and artificial intelligence.
What regulatory issues will legislators face?
Robots and artificial intelligence raise countless potential issues for regulators. Here are a few:
Safety standards. Regulators will have to know when a drone or driverless car is safe enough
to be permitted to transport people or goods. The prospect of emergent learning behaviour
will make this task particularly challenging.
Certification. Robots are beginning to develop the capability to perform tasks otherwise
reserved for certified professionals. For example, several teams are working on autonomous
surgical robots. Where and by what standard should such devices be certified?
Liability. In many instances, liability for robots will be straightforward. But there will be
circumstances where ordinary legal principles break down. For example, German courts or
lawmakers will have to decide whether an open robotic platform should be liable for third
party applications the user chooses to run on the robot. Also, the prospect of emergent
behaviour renders concepts such as foreseeability (in tort) and mens rea (in criminal law) less
useful—at least in American law.
Intellectual property. Patent law must attempt to strike a balance between fostering
collaboration in robotics and incentivizing innovation, as with any technology. But artificial
intelligence presents the further prospect that a system will either generate or infringe upon
intellectual property.
Privacy. Robots and artificial intelligence generate novel privacy issues. For example, artificial
intelligence may prove capable of making predictions and guesses about individuals or groups
that fall outside traditional notions of gathering personal data. Moreover, because we are
hardwired to react to anthropomorphic technology as though it were a person, citizens may
have fewer opportunities for solitude as interfaces become more social.
Taxation. One plausible scenario says that robots and artificial intelligence will displace jobs
in the short run and create jobs in the long run. If so, regulators may need to plan for a
revenue gap in terms of income taxes.
Apart from legal questions, for example concerning responsibility and liability, where will ethical
questions, in particular, also arise with regard to the use of artificial intelligence or as a result of the
aggregation of information and algorithms?
Robots and artificial intelligence raise just as many ethical questions as legal ones. We might ask, for
instance, what sorts of activities we can ethically outsource to machines. Does Germany want to be a
society that relegates the use of force, the education of children, or eldercare to robots? There are
also serious challenges around the use of artificial intelligence to make material decisions about
citizens in terms of minimizing bias and providing for transparency and accountability—issues already
recognized to an extent by the EU Data Directive.
2. What expectations do you have of this development – do you expect one or several industrial
revolutions – and if so, of what kind(s)? Might it even be possible to talk of a cultural revolution in
this connection? Can comparisons be drawn with previous industrial revolutions?
I lack the expertise to address this question.
3. How do you view proposals for the appointment of an (independent) expert or ethics commission
to answer social, legal and ethical questions, and potentially put forward concrete recommendations
for action?
I have argued for the establishment of a Federal Robotics Commission in the United States because of
my perception that U.S. regulators lack sufficient expertise in robotics and artificial intelligence to
generate an adequate law and policy framework. Independently, I have argued that companies
should establish ethical review boards for the research they conduct on consumers—including
through the use of machine learning.
For example, is it necessary to impose restrictions on the deployment of algorithms or artificial
intelligence for forecasting and concrete decisions, e.g. in the context of what is known as ‘predictive
policing’? In your opinion, how will it be possible to ensure the non-discriminatory neutrality and
transparency of automated decision-making processes? According to what principles will algorithms
– self-learning algorithms, in particular – define decisions as correct or incorrect?
Decisions about people using algorithms could, in theory, have greater accuracy and less bias than
decisions made entirely by people. The issue is that bias is very difficult to eliminate, in part because
the designers of algorithms do not understand their own biases, and in part because training data can
itself be biased. The German government could encourage the development of artificial intelligence
that is transparent and accountable, i.e., that presents not just recommendations but reasons. The
trouble is that many of today’s most promising machine learning techniques—neural nets, for
instance—do not admit of scrutiny in this way. Thus, testing for disparate impact on the vulnerable
and providing for opportunities for citizens to challenge algorithmically determined results becomes
4. It has been discussed over and over again in the course of digitisation whether jobs will fall victim
to automation and robotics or whether new jobs can be created. What studies are you aware of, and
how would you assess them? In what key technologies will robots replace human labour? Where do
you see opportunities and options for the creation of new jobs in Germany and Europe, especially as
a – direct or indirect – result of robotics and automation? Can the experience gained in other
countries be applied seamlessly to Germany? What do you see as the advantages of the
developments that are taking place – not only from economic points of view, but also as far as the
design of modern workplaces, more flexible working time models and/or a better work/life balance
are concerned?
I lack the expertise to address this question but recommend the book The Second Machine Age by Erik
Brynjolfsson and Andrew McAffee at MIT.
5. What sort of role will human beings have in the future world of work? Will there be a blurring of
the boundaries between the human and the machine?
For the foreseeable future, I hope and expect that artificial intelligence will mostly help humans make
decisions rather than replace human decision-makers entirely. This arrangement is advantageous
insofar as human-machine teams tend to outperform humans or machines alone. The challenge
becomes training people who work with artificially intelligent systems to understand them as tools
and not oracles. Robots will also help people perform physical tasks more safely and efficiently.
If so, in what fields will this happen and/or where is it already happening today?
The blurring of lines between person and instrument has been happening for decades in any number
of fields. Among other issues, American courts have had to decide whether robots represent
something “animate” for purposes of import tariffs, whether robots can “perform” as that term is
understood in the context of a state tax on performance halls, and whether a salvage team
“possesses” a shipwreck it visits with an unmanned submarine at maritime law. Moreover, many
cotemporary systems—from airplanes to cars—blend human and machine control for safety reasons.
Where do you see ethical dilemmas? What solutions can you propose to overcome these dilemmas,
where necessary?
An important ethical dilemma involves accountability. This problem runs both ways: We can imagine
human decision-makers or actors “blaming” the machine for any mistakes. Conversely, we can
imagine systems designed to keep a human being in the loop for the sole purpose of absorbing blame
or liability should something go wrong—what M.C. Elish has labelled a “moral crumple zone.”
6. What role is Germany playing in the development of robotics and artificial intelligence? Is
Germany competitive in international comparisons? What measures do you feel are necessary in
order to ensure continuing success? What is required, from your point of view, if German and
European start-ups in the field of robotics and automation are to be able to stand at the forefront of
global competition?
One way the German government can foster innovation in robotics is to resolve the question of
whether the manufacturer of an open robotic platform is liable when a user runs third party
applications on that platform. As with personal computers or smart phones, personal robots will be
more useful if they are sufficiently open to third-party contributions of software and hardware. Just as
you don’t want Apple to have to write every app for the iPhone, you don’t want a robot company to
have to write every app for its home robot. But open robots also present the potential for inestimable
legal liability, which may lead entrepreneurs and investors to abandon open robots in favour of
products with more limited functionality so as to avoid liability for what users do with robots.
7. How do you view the creation of common standards at the European and international levels?
From your perspective, is there a need for far-reaching amendments to legislation, e.g. in the field of
labour law, and will such legislative processes potentially be accompanied by the lowering of existing
standards of protection? What regulatory challenges will have to be coped with to ensure secure,
efficient flows of data traffic for networked, automated systems?
I know too little about European law to answer this question well. I will say that, from my perspective,
Europe has in place a robust data protection framework that will be relatively well-suited to address
the privacy challenges posed by robots and artificial intelligence.
8. The development of artificial intelligence, in particular, is playing a major role in the discussion
about robotics. What is your attitude to artificial intelligence?
The field of artificial intelligence is advancing rapidly, but narrowly. Many of the most promising
techniques were developed in the 1970s. Suddenly, due to sufficiently low storage costs for data and
access to greater and greater computing power, these same techniques yield astonishing results. For
example, Skype can now provide real-time language translation and IBM’s Watson can defeat our
greatest trivia masters. Some systems can even learn to accomplish tasks in surprising ways. But
artificial intelligence—and hence robots—remain fundamentally incapable of versatile problem
solving, let alone forming intent or exercising judgment.
In future, will robots also be able to perform emotional tasks that enable them to replace human
beings, e.g. in the care sector, road traffic or health services?
Robots can certainly detect and mimic emotions, which will make them more useful in a variety of
settings. Note, however, that detecting and mimicking emotions does not in any way require robots
to experience or understand them.
How will it be ensured that ethical criteria are complied with? Who will define these criteria, and can
you give examples?
Robot ethics begins with designers. The people that make robots and design artificially intelligent
systems should use existing methods such as value sensitive design (Batya Friedman) and responsible
innovation (Jeroen van den Hoven) to attempt to anticipate and address stakeholder concerns. I also
believe ethical review boards such as that of Deep Mind have a potential role—if conducted in a
meaningful and transparent way.
This being said, I believe government can and should step in to police against abusive or manipulative
behaviours. The fact that robots feel like people opens the door to questionable consumer practices.
The performance of such tasks would also involve value judgements and decision-making processes.
Is it conceivable that automated systems and artificial intelligence will be able to judge situations and
take decisions just as individually and situatively as human beings do? – Or even better?
There is a tension here. On the one hand, machines may be better able to make accurate and
unbiased determinations in many circumstances. On the other hand, machines cannot display
judgment, discretion, or compassion the way a person would. For the foreseeable future, artificially
intelligent systems should be thought of as tools and not oracles. This has several advantages. First,
research suggests that human-machine teams perform better than humans or machines alone.
Second, where people deviated from machine recommendations, we can reflect on why: Did they
deviate because of bias or because of judgment or compassion?
9. Automation, robotics and artificial intelligence will increase rationalisation and boost productivity.
For a long time, there have been proposals to introduce an ‘automation dividend’, ‘robot levy’ or
‘robot tax’, partly with a view to the future organisation of social security systems. Would such a levy
be worthwhile or necessary in your opinion?
In 1984, in the United States, a petitioner formally requested that the Internal Revenue Service to
impose a tax on companies that replaced workers with machines. The IRS declined, but there is much
to recommend a modest tax. First, it could control activity levels by forcing firms to think carefully
about whether the replacement of humans with robots makes fiscal sense. Second, the revenue
generated could fund retraining for displaced workers or otherwise provide a cushion for the
transition toward an economy with vastly greater automation. To be clear, however, I do not envision
a world in which every human retires in favour of robots.
10. Apart from specific questions concerning liability, fundamental legal questions are increasingly
being posed. In this respect, the central issue will be how decisions or actions that have been taken
by robots or ‘artificial intelligence’ and caused damage can be contested – potentially by means of
legal action. What fields of law are particularly affected by these issues, and what reforms would you
My own study of robotics law and policy suggests three general categories:
Embodiment. Robots are software that can touch you. The law will have to strike a different
balance now that bones instead of bits are at issue.
Emergence. Robots are increasingly capable of useful but surprising behaviour. Many areas of
law rely upon concepts such as foreseeability (tort), sufficient or exclusive control (tort,
agency, maritime law), and intent (criminal law, copyright) that the prospect of emergent
behaviour complicates.
Social valence. To a greater degree, perhaps, than any previous technology, robots blur the
line between person and instrument. We are hardwired to anthropomorphise them. Courts
have already had to draw lines between robots and people—an issue that will only amplify
with time, with implications for ethics, privacy, liability and other areas.
11. Against the background of current trends in research and development, please sketch out briefly
the direction in which, in your opinion, robotics and artificial intelligence will, or could, develop?
I believe robotics and artificial intelligence will continue to get better at domain specific tasks, and
even to surprise us with new capabilities. I also believe we will see greater innovation as we open op
robotic platforms to third party developers and to creative users. I doubt any field will remain
untouched by artificial intelligence and robotics—these tools are simply too useful. But I do not
expect, based on current trends, that the technology will approximate let alone surpass human
intelligence in my lifetime.
Thank you for this opportunity to address the Committee. I am honoured and look forward to a great

Report this document