PDF Archive

Easily share your PDF documents with your contacts, on the Web and Social Networks.

Share a file Manage my documents Convert Recover PDF Search Help Contact

Ethics for the machines Altitude Software .pdf

Original filename: Ethics for the machines - Altitude Software.pdf

This PDF 1.4 document has been generated by Chromium / ConvertApi , and has been sent on pdf-archive.com on 05/04/2018 at 09:22, from IP address 171.60.x.x. The current document download page has been viewed 194 times.
File size: 1.9 MB (8 pages).
Privacy: public file

Download original PDF file

Document preview


Ethics for the machines
Will the machines save us or kill us all? – that is the question. While many are thrilled with the latest AI
breakthroughs and dream of a shinning AI-powered world, others, like Bill Gates, Elon Musk, Steve Wozniak and the late
and legendary Stephen Hawking, expressed concerns about the evolution of the machines and warned about an
apocalyptic future.
But what do the machines tell us? In these days there are dozens of chatbots online that can understand us, right? So
I’ve asked the machines directly.

I don’t know about the future, but either the machines are hiding something very well or, except for some crude
nonsense, there is little to worry about in the present.
Although the answers are silly, all these chatbots are success cases, each one in its own domain, namely talk-therapy,
news and health. They were simply not programmed to give answers, or learn, outside their domain of knowledge.
That’s called narrow AI, or weak AI: arti cial intelligence applied to a very speci c goal.
Up to now all the AI successes have been “weak AI” successes (which doesn’t mean “weak successes” by any chance,
just limited to specialized tasks). Modern chatbots use AI mainly to extract relevant content from the user input in order
to select the best answer, and in some cases also for speech recognition purposes. A few other examples of
specialized AI applications are:
 Google Deepmind’s AlphaGo algorithm that has beaten Lee Sedol, the 18-time world champion of Go, in 2016.

Facebook face recognition, used to alert the user when a photo with him/her in it is posted by somebody else.
The DeepFace technology approaches human performance.
Net ix recommendation engine, with a ROI valued at £1 billion a year.
Uber’s machine learning platform, Michelangelo, trained with data from millions of trips to accurately estimate
arrival times and pick-up locations.
Google Maps automatic extraction of information from geo-located imagery. To increase the accuracy of search
results, a deep learning system analyzes over 80 billion photos taken by Street View cars to identify street names,
house numbers and business names from store fronts.
The recent Microsoft breakthrough in machine translation: a system capable of translating news articles from
Chinese to English with the same quality and accuracy as a person.
Tesla autopilot and self-driving cars. Tesla cars collect data from their sensing systems and sent it to the cloud so
the machine learning algorithms can learn and improve. According to Tesla’s CEO, Elon Musk, “the whole Tesla
eet operates as a network. When one car learns something, they all learn it”.

If we had a time machine, we wouldn’t need to travel back many years to astound everybody with this Tesla self-driving
Despite the amazing breakthroughs of the last 5 or 6 years, we can easily see that the current AI algorithms are too
narrow, or focused, to start understanding a wide range of domains or breed a conscience. For example, it is not
possible for a sophisticated Tesla self-driving car to learn Tic-tac-toe or any other trivial thing we can teach to human
child, neither suddenly become self-aware like Knight Riders’ KITT appeared to be. What we should hope for (or be
afraid of) is…

The technological singularity
The “singularity” metaphor was borrowed from theoretical physics, denoting a point in space and time where the
gravitational eld becomes in nite – for example at the center of a black hole. It means an event horizon that is hard to
see beyond, a point of no return. The technological singularity is the hypothesis that a super AI, after reaching humanlevel performance, will trigger an unstoppable technological growth by entering a “runaway reaction” of selfimprovement cycles, creating more intelligent generations of machines one after the other, resulting in a
superintelligence that will surpass by far all human intelligence, with an unpredictable outcome.

I. J. Good, a British mathematician who worked as a cryptologist with Alan Turing, introduced in the 60s the concept of
“intelligence explosion”. He wrote: “Let an ultra-intelligent machine be de ned as a machine that can far surpass all the
intellectual activities of any man however clever. Since the design of machines is one of these intellectual activities, an
ultra-intelligent machine could design even better machines; there would then unquestionably be an "intelligence
explosion," and the intelligence of man would be left far behind. Thus the rst ultra-intelligent machine is the last
invention that man need ever make, provided that the machine is docile enough to tell us how to keep it under control”. 
Not many were paying much attention to these ideas in the 60s, at least outside the realm of science ction (as a matter
of fact, I. J. Good served as consultant in Stanley Kubrick’s lm “2001: A Space Odyssey”). But given the recent advances
in the AI eld, the singularity has moved from science ction to serious debate in the last years. Ray Kurzweil, Director of
Engineering at Google but also an award-winning inventor and futurist, author of the bestseller “The Singularity is Near”,
predicts that it will happen by 2045 (hey, not far from the sci- future of “Blade Runner 2049”!). There is even a
movement for it, the Singularitarianism, and a think tank called “Singularity University”, founded in 2008 by Peter
Diamandis and Ray Kurzweil at the NASA Research Park.

The foundations of a new intelligence
The technological evolution in the last decades has allowed narrow AI to ourish, with some important breakthroughs
achieved in the last 6 years. That’s where machine intelligence is at the time: in the 2nd oor of the pyramid. If it reaches
the 3rd oor it will have capacities similar to the ones we nd in a human brain. It will be able to judge by itself whatever

input it gets and learn to do things is was not programmed for. This is called Arti cial General Intelligence (AGI), or
Strong AI. The 3rd oor does not seem near, but if the world keeps betting on AI, and it seems it will, one or two big
breakthroughs can change the game.
Once at AGI level, if a machine starts building improved copies of itself, the top of the pyramid, the singularity, will be
revealed probably fast, because the speed of thinking and development will be far greater than human speed, and
because it will have all the conditions to evolve: plenty of data to learn from, huge computing power, a world with large
machine networks and lots of humans that love technology and have computers everywhere: in their pockets, wrists,
cars, home appliances… oh, and also in their businesses, nance system, government o ces and military weapons.

Humanity loves gadgets - chart from Yahoo! Finance
What happens then? Will this super AI nd the cure for all diseases, discover the answers for the biggest questions,
create advanced technologies that will improve and extend the life of all humans – or will it try to exterminate mankind
to rule the world, like Terminator’s Skynet? Maybe none, maybe it depends on its main purpose. All we know is that it
doesn’t need to be an “AI super-evil mastermind” to cause massive damage: a bad designed paperclip maximizer is just
enough to do the job.

The Paperclip Maximizer
One thing we should realize is that, like human beings, super AI programs will want to achieve something. Nothing new,
narrow AI programs as Chess arti cial players or image recognition systems are always targeting speci c goals, but
super AI will be much more capable of creating strategies to avoid failure, including the ability of reprogram and
improve itself and create enhanced clones spread over the Internet. It will use any resources at its disposal with a
superhuman level of intelligence. If we don’t nd a way to program empathy, moral or ethics in the super AI software of
the future and understand how to create what Eliezer Yudkowsky calls “Friendly AI”, we can be in trouble.
To illustrate this idea Nick Bostrom, Director of the Oxford’s Future of Humanity Institute, created a surrealistic parable
called “The Paperclip Maximizer”.


Image borrowed from the “Universal Paperclips” game (of course there is one)
Imagine an AGI program created to run a paperclip factory with the goal of producing as many paperclips as possible.
At rst it does what it was programmed for: run the factory. But at some point, as its capabilities increase, an intelligence
explosion happens and it starts searching for better strategies to optimize the production. Now one factory is not
enough, it tries to get more factories and gather any possible resources, prevents anybody from switching it o by any
means, invents new technologies to produce paperclips and eventually takes over the world and colonizes other
planets for mass production of paperclips.
Although absurd, it illustrates the danger of advanced arti cial intelligence simply trying to accomplish the goal it was
programmed for, and, without a moral to tell right from wrong, creating sub-goals that con ict with ours. In fact, the
message of the parable remains intact if we replace the paperclips by anything else - contact center software, for
instance. Ok, maybe it would not have to end so dramatically, maybe this super smart AI would not enslave all mankind
and convert the solar system into a gigantic contact center before someone pulled the plug, but it would cause big
damage for sure and that’s at least a serious risk to consider.
Machines do not come with regulations, values, empathy, moral or ethics built-in. This quotation from Yudkowsky
(talking about a hypothetical super AI with molecular nanotechnology know-how) summarizes it well:

Eliezer Yudkowsky, Arti cial Intelligence as a Positive and Negative Factor in Global Risk

Machine ethics
The well-known science ction writer Isaac Asimov devised, in 1942, a set of rules for the robots called “The Three Laws
of Robotics”:
1. A robot may not injure a human being or, through inaction, allow a human being to come to harm.
2. A robot must obey orders given it by human beings except where such orders would con ict with the First Law.
3. A robot must protect its own existence as long as such protection does not con ict with the First or Second Law.

Let’s have fun with this while we can.
Source: https://xkcd.com/1613/ 
In the 40s, unless you were a gifted visionary, it made no sense to have a serious debate on the topic because there
were no smart machines in the horizon. Conversely, some decades later the debate not only started to make sense but
also became imperative, as the AI evolved and the machines began to make decisions.
Military semi-autonomous weapons are today’s reality. As technology evolves and robots become smarter, one day
they may carry orders autonomously and inevitably end up making life-and-death decisions in unpredictable scenarios.
At a civilian level, the same can be said about self-driven cars, for example. When they become common in our roads
they will be presented with the same ethical dilemmas the human drivers are, and they will have to decide. And then
comes the future of the future: the possibility of a singularity event raises ethical questions to a new alert level, as we’ve
Ok, what are we doing about it? First of all, thinking. Asimov’s laws were naïve, but visionary, as they in uenced an
emerging and important sub- eld of philosophy called machine ethics.
Machine ethics is concerned with giving AI machines ethical principles, or procedures for nding a way to resolve
ethical dilemmas, enabling them to act in an ethically responsible manner through their own ethical decision making.
This new eld is therefore interdisciplinary, evolving not only philosophers but also AI researchers towards the goal of
nding how an ethical dimension can be added to the smart, autonomous machines of the future.
If one side of the discussion applies mainly to governments and global organizations on how to regulate the use and
the production of autonomous robots, the other side, concerning the development of a friendly, or ethical, AI before a
singularity event, applies to companies, universities and all kinds of institutions involved in AI research. In fact, given the
worst-case scenario, it may apply to everyone on the planet.
Some alerts have already been issued. In 2015 over a thousand AI researchers signed an open letter urging the United
Nations to ban the development and use of autonomous weapons. The letter was released by the Future of Life
Institute. Another group working for the same goal is the “Campaign to Stop Killer Robots”, co-founded by Human
Rights Watch.
That makes us go back to the beginning of this post, as the question remains: are the machines going to save us or kill
us all? Finishing with one last try:

First Name *

Last Name

Email *


Comment *

Subscribe to follow-up comments for this post
protected by reCAPTCHA
Privacy - Terms


Share this:




Most popular
Can a Contact Center be a “Deep Learner”?
Are you GDPR Ready? 10 Questions You Should Be Asking
Winning Hearts and Minds: Five Ways Behavioral Pairing Improves Contact Center Performance


Subscribe to Email Updates

Email *




Related documents

ethics for the machines altitude software
geek out 2017 usa
1 s2 0 s004016251630539x main
tori ver 6

Related keywords