A good overview of Artificial Intelligence. If you know nothing about it, start here, then read “Life 3.0” afterwards.
Singularity means a point where the normal rules cease to apply, and what lies beyond is un-knowable to anyone this side of the event horizon.
Intelligence measures an agent’s general ability to achieve goals in a wide range of environments.
Nine types of intelligence:
Whether intelligence resides in the machine or in the software is analogous to the question of whether it resides in the neurons in your brain or in the electrochemical signals that they transmit and receive.
Two very different types of artificial intelligence:
artificial narrow intelligence (ANI) - also known as weak AI
artificial general intelligence (AGI) - also known as strong or full AI
An AGI is an AI system which can carry out any cognitive function that a human can.
We can learn useful lessons from one activity and apply them to another.
A machine that learns snakes and ladders has to forget that in order to learn how to play Ludo.
This is called “catastrophic forgetting”.
Cars are not called artificial horses.
Planes are not called artificial birds.
A machine learning algorithm uses an initial data set to build an internal model which it uses to make predictions. It tests these predictions against additional data and uses the results to refine the model.
90% of all the data in existence was created in the last two years.
Turning Big Data into information and thence into understanding and insight is the job of algorithms.
Having more data beats having better data if what you want is to be able to understand, predict and influence the behaviour of large numbers of people.
Traditionally, statisticians started with a hypothesis and went to look for data to support or refute it.
Machine learning, by contrast, starts with data – lots of it – and looks for patterns.
Backpropagation retraces the steps in a neural net to identify the errors made in an initial processing of input data.
Categorize machine learning researchers into five tribes: the symbolists, the evolutionaries, the Bayesians, the analogisers, and the connectionists.
Symbolists = Good Old-Fashioned AI.
Evolutionists start with a group of possible solutions to a problem and produce a second “generation” by introducing small random changes and removing the least effective solutions.
Bayesians make hypotheses about uncertain situations and updates its degree of belief in each hypothesis according to a mathematical rule when new evidence is provided.
Analogisers: new phenomenon is likely to follow the same behaviour as the previously observed phenomenon which is most like it.
Connectionists are trying to reverse engineer the brain by developing artificial neural networks.
Connectionism has been re-branded as deep learning, and it has been the most successful form of machine learning so far.
Convolutional networks: the neurons in each layer are only connected to groups of neurons in the next layer.
Recurrent networks: connections from each layer can loop back to an earlier layer.
Supermarkets use algorithms to predict what we will collectively want to buy, when and where.
Hospitals use AI to allocate beds.
Telecoms companies, power generators and other utilities use it to manage the load on their resources.
Google's ambitions are startling, and the hurdle for potential acquisition targets is high: they must pass a “toothbrush test”, meaning that their services must be potentially useful to most people once or twice every day.
MemNet predicts the memorability of images.
Art: the use of a craft skill to communicate something profound or at least interesting about the condition of being human.
Amara’s Law: we tend to over-estimate the effect of a technology in the short run and under-estimate the effect in the long run.
You are composed of around 27 trillion cells, which were created by fission, or division – an exponential process.
It required 46 steps of fission to create all of your cells.
While classical computers use bits (binary digits) which are either on or off, quantum bits (qubits) can be both on and off at the same time – known as superposition. This enables them to carry out a number of different calculations at once.
The hedonic treadmill: we quickly become accustomed to the new reality, and what seemed wonderful in prospects becomes ordinary.
Anything that is in the world when you’re born is just a natural part of the way the world works.
Anything invented between when you’re 15 and 35 is new and exciting.
Anything invented after you're 35 is against the natural order of things, and should be banned.
Most of the tasks that we perform each day can be broken down into four fundamental skills:
Restorative justice, in which criminals do as much as possible to undo the harm they have done, is better all round than purely retributive justice.
In math and physics, the term “singularity” means a point at which a variable becomes infinite.
The classic example is the centre of a black hole, where the gravitational field becomes infinite, and the laws of physics cease to operate.
The three biggest questions about artificial general intelligence (AGI) are:
1. Can we build one?
2. If so, when?
3. Will it be safe?
The first of these questions is the closest to having an answer, and that answer is “probably, as long as we don't go extinct first”.
Our brains were developed by a powerful but inefficient process called evolution.
The apparent consciousness of the octopus, with which we share no recent common ancestor, suggests consciousness is more than coincidence.
For ages, we tried to fly by copying what birds do, but when we finally did learn to fly, it was not by copying birds.
There are still things we don't understand about how birds fly, yet we can now fly further and faster than they can.
I don’t worry about turning AI evil now for the same reason I don't worry about the problem of overpopulation on the planet Mars.
Artificial superintelligence (ASI) is generally known simply as superintelligence. It does not need the prefix “artificial” since there is no natural predecessor.
Neurons signals in our brain travel around 100 metres per second.
Synapse crossing is chemicals jumping across the gap.
Synapse jumping part is much slower than the electrical part.
Signals within computers typically travel at 200 million metres per second.
So a brain emulation AGI could operate 2 million times faster than a human.
When we create superintelligence, will we create one, two, or many?
The first superintelligence may take steps to prevent the creation of a competitor.
Superintelligence with human levels of cognitive ability would be a technological marvel, and a harbinger of things to come.
Superintelligence would be the game-changer.
Our future will depend on its decisions and its actions.
Transhumanists believe we have every right to enhance our physical and mental selves using whatever technology we can create – perhaps even that we have a moral duty to do so.
Humanity versus a fully fledged superintelligence with internet access would be like the Amish versus the US Army.
Intelligence brings enlightenment and enlightenment brings benevolence.
The Whig view of history sees the human story as an inevitable progression towards ever greater enlightenment and liberty.
We are not intrinsically more violent than other animals. No other carnivore is capable of assembling more than a dozen or so members in one location without a serious fight breaking out.
We can use AI to augment us rather than having to compete with it. This is called Intelligence Augmentation, or IA, and is also known as intelligence amplification.
The best chess players today are neither humans nor computers, but a combination of both — labelled “centaurs” by Gary Kasparov.
Musk believes that humans cannot afford to let machines get much smarter than them, and the only way to avoid that is to – at least partly – merge with them.
Asimov's Three Laws of Robotics:
How far into the future would a robot have to project the consequences of an individual action?
How would it assign probabilities to the various different possible outcomes?
Since there should be no time limit, a robot would be rendered inactive by the amount of calculation required prior to any action.
Humans – as well as the superintelligence – will evolve over time and would not want to be stuck with the particular ideas of morality and pragmatism which happened to prevail in the 21st century.
Few of us even know the names of Vasili Arkhipov and Stanislav Petrov, two men who quite literally saved the world during the Cuban missile crisis.