CS - Ethics - Presentation

Author: Francesca Rossi EN Policy Department C: Citizens' Rights and Constitutional Affairs European Parliament PE 571.380

Artificial Intelligence: Potential Benefits and

Ethical Considerations

KEY FINDINGS

 The ability of AI systems to transform vast amounts of complex, ambiguous

information into insight has the potential to reveal long-held secrets and help solve

some of the world’s most enduring problems.

 However, like all powerful technologies, great care must be taken in its development

and deployment. To reap the societal benefits of AI systems, we will first need to trust

them and make sure that they follow the same ethical principles, moral values,

professional codes, and social norms that we humans would follow in the same

scenario. Research and educational efforts, as well as carefully designed regulations,

must be put in place to achieve this goal.

 International Business Machines Corporation (IBM) is actively engaged, both internally

as well as with its collaborators and competitors, in global discussions about how to

make AI ethical and as beneficial as possible for people as society.

1. WHAT IS ARTIFICIAL INTELLIGENCE?

The term “artificial intelligence” (AI) has been mentioned for the first time in 1956 by John

McCarthy during a conference where several scientists decided to meet to see if machines could

be made intelligent. Since then, AI is usually defined as the capability of a computer

program to perform tasks or reasoning processes that we usually associate to intelligence

in a human being. Often it has to do with the ability to make a good decision even when there

is uncertainty or vagueness, or too much information to handle.

As an example, playing chess well, or some complex card games, is believed to need some

form of intelligence in a human being, as well as choosing the best diagnosis in a difficult

medical case, or creating something new, such as a mathematical theorem or even some form

of art, or even driving a car in the middle of a crowded city.

It is clear that this is a strange definition, because it depends on what we consider being

intelligent in the behaviour of a human being at a certain point in time. If our belief about

human intelligence changes, and we don't believe any longer that a certain task requires

intelligence, then a computer program performing that task is no longer part of AI, it becomes

just another boring computer program.

The term “artificial intelligence” brings to mind to the notion of replacing human intelligence

with something synthetic. At IBM, we prefer the term “augmented intelligence”. This means

that we aim to build systems that enhance and scale human expertise and skills rather than

replacing them. We therefore focus on practical applications of discrete AI capabilities that

assist people in performing well-defined tasks, by exploiting a wide range of AI-based services.

We also use the term “Cognitive Computing”, to mean a comprehensive set of capabilities

based on technologies which include AI, but that go far beyond it. “Cognitive Computing”

Policy Department C: Citizens' Rights and Constitutional Affairs

2 PE 571.380

comprises the fields of machine learning, reasoning and decision technologies, language,

speech and vision recognition and processing technologies, human interface technologies,

distributed and high-performance computing, and new computing architectures
 and devices. When purposefully integrated, these capabilities are designed to solve a wide range of practical

problems, boost productivity, and foster new discoveries across many industries.

2. AI IN OUR LIVES

There are many examples of the presence of AI in our current life, that we don’t even know of.

Whenever we buy something with a credit card, an AI algorithm approves that transaction (or

not). When we use the GPS in our car, the algorithm that finds the best way to go from where

we are to where we need to go is called the A* algorithm and it is an essential tool for AI,

present in every AI teaching book. Spam filters are based on AI. Recommender systems such

as that of Amazon are AI. The Google translate service, which nowadays is able to translate

from and to more than 70 languages, is based on statistical machine learning, which is part of

AI. Even web searches, such as those that we ask of Google or Baidu or other search engines,

rely on AI to give us the web pages that are most relevant to our query. The face recognition

capability of any of our cameras, shown usually with a green rectangle around each face we

want to take a picture of, is AI. Siri, the IPhone app that understands us when we speak and

responds (usually) in a useful way, is based on AI algorithms for speech understanding.

And of course there is the whole branch of robotics, which is more easily associated with AI

because of the iconic image of humanoid robots that make it seem that humans have been

reproduced artificially. Of course not all of them are intelligent in the way we would say a

human is intelligent, but they are usually very good at doing what they are supposed to in their

environment, from the Roomba robot that cleans the floors of our houses, to the Baxter robot

that can work together with humans in production chains, passing through the Kiva warehouse

robots that can take care of the tasks of an entire warehouse and the companion robots like

Nao, Pepper, Aibo, and Giraff, who can entertain us, talk to us, and help elderly people to stay

connected to their friends, relatives, and doctors.

The realm of possible uses of AI techniques is enormously vast, and this is one of the reasons

why many companies have been heavily investing in AI in recent years. Google is building self-

driving cars and has acquired more than 10 robotics companies, Facebook had opened a whole

new research facility focused only on AI research, Apple has developed Siri, Microsoft has built

Cortana, a similar personalized assistant, Google has acquired DeepMind, a UK company whose

long-term aim is to build general AI and has already shown great potential in winning at the

game of Go the current world champion, and IBM is investing a huge amount of resources in

applying its Watson cognitive computing system to the medical domain, to finance, and to

personalized education, just to name a few. This expansion of AI-based systems and services

is reaching all corners of the globe. In Europe, IBM is establishing new centres in Munich and

Milan focused on the application of cognitive computing capabilities to the Internet of Things

and healthcare, respectively.

Self-driving cars are all about AI: they need to be able to see what happens in the street

(signals, lanes, other cars, pedestrians, traffic lights), they need to able to predict what other

cars and pedestrian will do, and they need to be able to cope with unforeseen situations. Since

most car accidents are due to human fault, it is estimated that the adoption of self-driving cars

will save about half of the lives that are usually lost in car accidents, which totals around 40,000

each year in the US alone. Some of us may be reluctant to hand over the wheel to an AI system,

but very soon we may wonder why we did not do it sooner!

Watson is an IBM cognitive computing system that won against the best human champions at

the Jeopardy! game in 2011. To do that, IBM Watson had to understand spoken language,

make sense of massive amount of text, respond correctly to questions in many categories, as

well as assess its own confidence in responding to such questions. The kind of

question/answering capabilities that would be very useful, for example, in assisting a doctor

Artificial Intelligence: Potential Benefits and Ethical Considerations

PE 571.380 3

when trying to come to the correct diagnosis for a patient and to propose the best therapy.

Watson puts together many AI results and algorithms, from text and speech understanding, to

reasoning with uncertainty, to optimization.

IBM is not new to tackling daunting challenges and successfully addressing them. In 1997, the

Deep Blue computer program won against the world chess champion Garry Kasparov. This was

very iconic, since chess, as I said above, is one of those activities that we believe requires a

significant amount of intelligence in a human being. Deep Blue showed that computers could

do better than the best humans when it comes to certain tasks.

3. AI AND COMPUTING POWER

We have to be careful in labelling all this promising progress as truly “intelligent.” Humans

need intelligence and good intuition in playing chess because our brain does not have enough

computing power to make sense of a lot of data. In chess, for example, it is not possible for

our brain to evaluate all possible sequences of moves of us and our opponent in a very short

time. If we could do that, it would be obvious to us what the best move is. Contrarily to our

brain, computers can rely on a computing power that, according to Moore’s law, doubles every

about 18 months. Gordon Moore, co-founder of Intel, in 1965 noticed that this was the trend

in putting transistors into a single chip, and to the amazement of many, this law has been

followed since then in our computers.

This law means that computer processing power doubles every 18, or, seen from another point

of view, every 18 months we can have a much cheaper computer with the same speed as the

old one. For example, it has been calculated that an IPhone in 1991 would have cost about

$3,6 million. And this is only for its processor, its memory, and its connectivity. Today’s smart

phones are more powerful than NASA computers that in 1969 sent a man to the moon. This is

how much computing power has increased over the years.

4. AI AND DATA

AI is not all about computing power. Intelligent machines can also rely on huge amounts of

data, to be used to learn how to make better and better decisions. This data comes from all of

us. Over the years, Facebook users have uploaded more than 250 billion pictures, and every

day they upload about 350 million more. Every second, we submit 40,000 Google search

queries, which means 3.5 billion per day and 1.2 trillion per year. As of today, there are 2

billion people connected to internet, which is estimated to get to 5 billion by 2020. And by that

time, also 50 billion “things” will be connected through the web: from appliances to traffic

lights, from cars to watches.

5. MACHINES VS HUMANS

No matter how much data and computing power is available to machines, there are tasks that

are still difficult for machines to perform, but that remain very easy for humans. Machines and

humans are very complementary. A typical example is understanding what is depicted in an

image.

How do we know that an image contains a cat? Because during our life we have seen many

examples of cats and non-cats, and at some point we got a very good idea of how a cat should

look like, so much that we don’t have problems recognizing one even if we have never seen it

before, and even if it is in a strange position.

Machines need humans to provide them with many examples. A lot of progress has been made

in this, but we are still working hard to improve their accuracy in labelling pictures or other

perception capabilities

Policy Department C: Citizens' Rights and Constitutional Affairs

4 PE 571.380

Other tasks that are very easy for humans are physical and manipulation tasks such as walking,

running, picking up an object no matter its shape and location. Robots can do this only in

restricted environments. But they are still not able to have the general physical and

manipulation capabilities even of a 6 year old.

6. AI ETHICS AND TRUST

The ability of AI systems to transform vast amounts of complex, ambiguous information into

insight has the potential to reveal long-held secrets and help solve some of the world’s most

enduring problems. AI systems can potentially be used to help discover insights to treat

disease, predict the weather, and manage the global economy. It is an undeniably powerful

tool. And like all powerful tools, great care must be taken in its development and deployment.

However, to reap the societal benefits of AI systems, we will first need to trust it. The right

level of trust will be earned through repeated experience, in the same way we learn to trust

that an ATM will register a deposit, or that an automobile will stop when the brake is applied.

Put simply, we trust things that behave as we expect them to.

Trust is built upon accountability. As such, the algorithms that underpin AI systems need to be

as transparent, or at least interpretable, as possible. In other words, they need to be able to

explain their behaviour in terms that humans can understand — from how they interpreted

their input to why they recommended a particular output. To do this, we recommend all AI

systems should include explanation-based collateral systems.

But trust will also require a system of best practices that can help guide the safe and ethical

management of AI systems including alignment with social norms and values; algorithmic

responsibility; compliance with existing legislation and policy; assurance of the integrity of the

data, algorithms and systems; and protection of privacy and personal information.

One of the primary reasons for including algorithmic accountability in any AI system is to

manage the potential for bias in the decision-making process. This is an important and valid

concern among those familiar with AI. Bias can be introduced both in the data sets that are

used to train an AI system, and by the algorithms that process that data. At IBM, we believe

that the biases of AI systems can not only be managed, but also that AI systems themselves

can help eliminate many of the biases that already exist in human decision-making models

today.

AI systems should function according to values that are aligned to those of humans, so that

they are accepted by our societies and by the environment in which they are intended to

function. This is essential not just in autonomous systems, but also in systems based on

human-machine collaboration, since value misalignment could preclude or impede effective

teamwork. It is not yet clear what values machines should use, and how to embed these values

into them. Several ethical theories, defined for humans, are being considered (deontic,

consequentialist, virtue, etc.) as well as the implications of their use within a machine, in order

to find the best way to define and adapt values from humans to machines.

In industries like healthcare and finance, the relevant professional ethical principles are

explicitly encoded and practiced by professionals in the field already. In AI systems designed

to help professionals in these domains, these best practices and principles could form the core

of the ethics module for such systems. Ethics modules, however, should be constantly adapted

to reflect humans’ best practices in their everyday profession.

We envision a future in which every AI system will need to have its own ethics module to allow

for a fruitful interaction and collaboration with humans
 in the environments in which it is used. This could be achieved by developing an ethics API that can be adapted to specific professions

and real-life scenarios. It would provide the main principles and values the AI systems should

base its behaviour on, as well as the capability to dynamically adapt them over time to tune

them to the real situations that are encountered in that profession or environment. Such a

Artificial Intelligence: Potential Benefits and Ethical Considerations

PE 571.380 5

rigorous approach could offer sufficient value alignment without compromising the full problem-

solving potential of artificial intelligence.

7. IBM AND AI

IBM has been researching, developing and investing in AI technology for more than 50 years.

In 1997, IBM Deep Blue bested then world chess champion Garry Kasparov, showing that

innovative AI algorithms and computational power can play a complex game at super-human

levels. In 2011, IBM Watson won at Jeopardy! against the best human players, showing that

AI can also perform very well in natural language understanding and reasoning with

uncertainty.

These are just the tip of the iceberg compared to what IBM has been achieving over the years

in the field of AI. We have been transforming the original Watson program into a fully fledged

platform and we have exploited it to successfully apply AI to many industrial sectors, including

healthcare, finance, commerce, education, security, and the Internet of Things. The whole

company is deeply committed to AI, since we believe strongly in its potential to benefit society

while transforming our personal and professional lives.

As mentioned above IBM prefer the term “augmented intelligence” and therefore focuses on

practical applications of AI capabilities that assist people in performing well-defined tasks, by

exploiting a wide range of AI-based services. With this aim in mind, IBM researchers, in tight

collaboration with several universities, produce continuous innovations in area such as machine

learning, knowledge modelling, reasoning and decision technologies, human interface,

automated perception, data assurance, and computing infrastructures. Most of these research

efforts cannot be achieved by AI researchers alone. Collaboration with experts in multiple

disciplines
 — such as psychology, philosophy, sociology, art, regulation, and law — is crucial.

We believe that new companies, new jobs, and entirely new markets will be built on the

shoulders of this powerful technology. Moreover, AI systems will improve access to critical

services for underserved populations. Overall, we anticipate widespread improvements in the

quality of life.

8. IBM AND AI ETHICS

In order to be fully accepted into society, AI systems need to have significant social capabilities,

because their presence in our lives has a profound impact on our emotions and on our decision

making capabilities. Also, AI systems need to understand how to learn and comply with specific

behavioural principles for aligning with human values. To take full advantage of the potential

societal benefits of AI, we will need to trust AI, whether we speak of autonomous systems or,

as is the focus of IBM, of human/machines partnerships. Trust will be earned over time and via

natural interaction modalities. Trust will also require a system of best practices that can guide

the safe and ethical development and management of AI, a carefully thought alignment with

social norms and values, algorithmic accountability, compliance with existing legislation and

policy, and protection of privacy and personal information.

IBM is in the process of developing this system internally, with our collaborators, and also with

our competitors. More precisely, IBM is engaged in several efforts –
 both internally and externally – to advance our understanding and effecting the ethical development of artificial

intelligence. They include:

 The establishment of an internal IBM Cognitive Ethics Board, to discuss, advise and

guide
 the ethical development and deployment of AI systems.  A company-wide educational curriculum on the ethical development of cognitive

technologies.

Policy Department C: Citizens' Rights and Constitutional Affairs

6 PE 571.380

 The creation of the IBM Cognitive Ethics and Society research program, a multi-

disciplinary research program for the ongoing exploration of responsible development of AI

systems aligned with our personal and professional values.

 Participation in cross-industry, government and scientific initiatives and events around

AI and ethics, such as the recently launched Partnership on AI, the White House Office of

Science and Technology Policy AI workshops, the International Joint Conference on Artificial

Intelligence, and the conference of the Association for the Advancement of Artificial

Intelligence.

 Regular, ongoing IBM-hosted engagements with a robust ecosystem of academics,

researchers, policymakers, NGOs and business leaders on the ethical implications of AI.

9. IBM AND EUROPE

IBM is an international company with a strong history and presence in Europe:

The IBM Zurich research lab is supported by a multicultural and interdisciplinary team of a few

hundred people from about 45 nationalities who work in diverse areas such as chip

technologies, nanotechnology, fibre optics, supercomputing, data storage, security and privacy,

risk and compliance, business optimization and transformation, and server systems. The Zurich

lab is involved in many joint projects with universities throughout Europe, in research programs

established by the European Union and the Swiss government, and in cooperation agreements

with research institutes of industrial partners.

The recently opened IBM Watson IoT Headquarters in Munich is applying Watson to the Internet

of Things and helping many companies (from automotive, insurance, electronics, banks, and

industrial sectors) to transform their business by extending the power of AI to the billions of

connected devices, sensors and systems that comprise the IoT.

The recently announced IBM Watson Health European Centre of Excellence, that will be placed

within the Human Technopole Lab in Milan, is supporting the government of Italy’s initiative to

establish an international hub for the advancement of genomics, big data, aging, and nutrition.

The Centre is expected to provide access to resources and technology designed to help

accelerate research into new treatment options, promote personalized medicine, and

encourage discoveries aimed at improving overall public health management while advancing

sustainable health systems.

10. AI AND POLICIES

AI technology is changing so rapidly, and has so many applications to the real world, that it is

difficult for any government or regulatory agency to keep up with them and to meaningfully

and timely guide the deployment of AI systems. However, some issues like data privacy and

ownership have been considered in the EU, as well as algorithm transparency and

accountability.

An example is the recently released General Data Protection Regulation, that will take effect as

law across the EU in 2018 and will restrict automated individual decision-making (that is,

algorithms that make decisions based on user-level predictors) which "significantly affect"

users. The law will effectively create a so-called "right to explanation," whereby a user can ask

for an explanation of an algorithmic decision that was made about them. Another example is

the very recently released USA federal policy on automated vehicles, that is already in effect.

The main point of all these policies is to make sure that society can take full advantage of the

capabilities of AI systems while minimizing the possible undesired consequences on people.

Safety is very important, as well as fairness, inclusiveness, and equality. These and other

properties should be assured of AI systems, or at least we should be able to assess the limits

of an intelligent machine, so to not overtrust it. It is therefore very important the policies and

regulations help society in using AI for the best of all.

Artificial Intelligence: Potential Benefits and Ethical Considerations

PE 571.380 7

Ethical issues, including safety constraints, are essential in this respect, since an AI system

that behaves according to our ethical principles and moral values would allow humans to

interact with it in a safe and meaningful way.

It is clear that a lack of regulations would open the way to unsafe developments. However, also

excessive regulations would have a cost to society, since they would not allow us to take

advantage of all the potential benefits that AI can bring, such as saving lives, curing diseases,

and solving planetary problems.

IBM is eager to work with governments, media, other companies, regulatory agencies, and

industry sectors in a meaningful discussion on ethical issues of AI, with the aim of clearly

identifying the potential and limits of AI, and carefully understanding how to harness it for the

best of all.

Policy Department C: Citizens' Rights and Constitutional Affairs