Skip to main content
Loading Explore guides…
AI Philosophy · Acedra
Home
Explore
Philosophy
AI Philosophy
Sign in to save
Chat
If YouTube asks you to sign in, open the video on YouTube.
Watch on YouTube
Reload player
Transcript
72 segments
Copy
Hide video
0:00
artificial intelligence will likely dramatically impact your life in the coming years but do you know what exactly it is and the major implications surrounding its development well most people don't so in this video we're going to be covering the philosophy of artificial intelligence the branch of philosophy that explores what artificial intelligence specifically is in other philosophical questions surrounding it like can a machine act intelligently is the human brain essentially a computer
0:25
can a machine be alive like a humanist Can it have a mind and Consciousness can we build Ai and align it with our values and ethics if so what ethical systems do we choose we're going to be covering all those questions and possible answers to them and what will hopefully be an easy to understand 101 style manner so what is artificial intelligence exactly to answer that question we need to understand what intelligence is first and if something artificial or a machine
0:52
can have it the term artificial intelligence was coined and the fyoto study officially launched in 1956 during a DARPA sponsored event at Dartmouth College however prior to this important event many individuals were already thinking about artificial intelligence related questions including the idea that machines could or couldn't be intelligent one of the first thinkers to write about whether machines could display human level intelligence in the first place was Renee Descartes
1:19
considered by some to be the father of modern philosophy basically he believed machines or automata could never replicate human intelligence writing different automata removing machines could be made by the industry of man for we can easily understand a machines being constituted so that it can utter words and even emit some responses to action on it of a Corporal kind which brings about a change in its organs for instance if touched in a particular part it may ask what we wish to say to it if
1:47
in another part it makes claim that it's being hurt and so on but it never happens that it arranges its speech in various ways in order to reply appropriately to everything that may be said in its presence as even the lowest type of man could do okay so was he right all along can a machine never exhibit real intelligence well first we have to Define intelligence something philosophers and linguists have struggled to do Alan Turing dubbed the father of computer science wondered if a
2:13
machine could overcome this deficiency of language and the supposed lack of intelligence Descartes described in a famous 1950 paper Computing machinery and intelligence he believes we should switch out the question can machines think with the question can a machine be linguistically indistinguishable from a human to help answer the intelligence question he thought intelligence is hard to Define but we know it when we see it it's manifested in our speech and when
2:37
we talk to another person we have a polite convention that they're intelligent so why not do that with a machine a computer would deserve to be called intelligent if it could deceive a human into believing it was human he wrote and then he goes on to devise what's known as the Turing test a way to prove a machine has human level intelligence the standard interpretation of the test goes as follows the woman in a computer or in different rooms and a human judge must
3:02
figure out which room contains the machine in which the woman and he asks questions both via email or in that day uh that's called teletype and if you could do no better than 50 50 when delivering the verdict as to which room the computer was in that computer has passed the Turing test it could also be said to have human level intelligence however there's been some criticisms of this way of defining an intelligent machine including that the test isn't measuring intelligence more of the
3:25
humanness of the machine in question human behavior and intelligent Behavior are not exactly the same thing after all so back to the first question what is artificial intelligence well there's no consensus definition still to this day however in one of the most comprehensive works on AI artificial intelligence a modern approach authors Rosso and Peter norvig have grouped all possible answers to this question into a simple chart where they assume a i is defined in
3:51
terms of its goals the goals of AI are centered around two questions so we judge AI based on its similarities to humans for example like Turing or an ideal rational being and should we judge AI based on its ability to think or its ability to act the definition of AI would be AI is the field that aims at building and then you insert your preferred position so what would uh Turin Turing think well Turing would be AI is the fuel that aims at Building Systems that act like humans
4:21
what do Steward and Norfolk believe well when we think of intelligence we think not of human behavior necessarily but of rationality Aeronautical Engineering text they wrote do not Define the goal of their field as making machines that fly so exactly like pigeons so that they can fool their pigeons they see AI as the field of study that aims at Building Systems that act rationally specifically what they call intelligent agents now that would put them in the systems that
4:47
act rationally camp now what's an intelligent agent explain that right at the beginning of the textbook the main unifying theme is the idea of an intelligent agent we can define AI as the study of agents that receive precepts or perceptions from the environment and perform actions each such agent implements a function that Maps precept sequences to actions and we cover different ways to represent these functions and then they give this example of what they mean by an
5:14
intelligent agent now in the book they outline how these agents can become increasingly complex so we have a simple looking reflex agent here but this can become more complex to something like where the agent is now self-learning but one could argue that something like a thermostat could potentially count as a rational agent and thus is AI but we don't think of things like thermostats as AI really now there's one definition outlined by one of the founding fathers of AI Alan Newall given that the
5:42
original Dartmouth conference which encapsulates all of Russell and norfolk's possibilities which is AI is the field devoted to building artifacts that are intelligent where intelligent is operationalized through intelligence tests such as the wechsler adult intelligence scale and other tests of mental ability including test of mechanical ability creativity and so on so we might not ever achieve a full consensus of what AI is but this definition might be satisfactory for now
6:08
okay since we understand AI by the goals of the field there are some important ways of categorizing it to help further understand what it is now there's two dichotomies we need to go over which are weak AI versus strong Ai and narrow artificial intelligence versus General artificial intelligence the philosopher John Cyril has proposed two goals of AI development to help us further understand and explore what it is and these are weak Ai and strong AI first week AI are informational processing
6:37
machines that appear to have all the mental capacities of living humans or animals but are not alive so they don't have conscious experience however a weak AI system can pass the Turing test or the total Turing test this is a concept where a machine passes for a human in all aspects such as throwing a ball and being creative strong AI is artificial persons or creatures that are alive meaning they have all the mental capacities including self-awareness and consciousness and to help distinguish
7:06
between the two serial rights according to strong AI the correct simulation really is a mind according to weak AI the correct stimulation is a model of the mind and he created these terms to help formulate an argument in which artificial intelligence executing a program can never have a mind and be conscious um and I'm going to be covering the argument more in depth and exactly what it means in just a minute but first note that these terms are not to be confused with narrow Ai and AGI so
7:34
let's cover those real quick now when we say narrow AI we mean the process of making a computer system that can do specific tasks or things that normally require human intelligence such as understanding natural language and recognizing objects the program can't do everything human intelligence can do just a specific aspect of human intelligence for examples of narrow AI or image recognition and facial recognition systems uh chat Boston conversational assistants self-driving vehicles
8:01
on the other hand AGR is a program or a machine that can understand and learn any intellectual task that a human being can and the performance of these systems is indistinguishable from that of a human the broad and electrical capacities of AGI would exceed human capacities because of its ability to access and process huge data sets at incredible speeds so AGI sets out to understand systems and perform them better than humans can while a narrow AI conserves itself with specific tasks or
8:29
problems then there's something called super intelligence this could be a goal of the AGI or a combination of several AGI and philosopher Nick Bostrom defines super intelligence as any intellect that greatly exceeds the cognitive performance of all humans in virtually all domains of interest and Bostrom trade superintelligence as general dominance at all goal-oriented Behavior becoming far superior in any activity than a human can do again an AGI application can rise to a super
8:59
intelligence level or several AGI can combine to form super intelligence and sometimes you may hear strong AI used interchangeably with AGI and Nai used interchangeably with weak Ai and this is not exactly correct it's theoretically possible to have AGI but it doesn't classify as seagull's strong AI okay so we have a general understanding of what AIS at this point so can a machine really display a human-like intelligence and perhaps more importantly can a machine be alive and
9:29
strong AI fully realized so Descartes said that it's not possible but I think most philosophers would disagree with him at this point at least when it comes to weak AI now today there doesn't seem to be any grounds to think that weak AI is in a philosophically sound idea it might be achieved in the next few years if you think about it the nervous system obeys the laws of nature as we currently understand them so if you can replicate the brain in the nervous system there's
9:54
no reason to think the artificial replicant won't produce the same behavior as the natural mind now in the philosophy of mine there's an idea that helps explain this called computationalism AR research and philosophy is drawing from computationalism states that the mind is a computational system that is physically implemented by neural activity in the brain cognition and Consciousness are a form of computationalism and the mind arises from computation in the brain now this
10:21
theory was proposed by Hillary Putnam and Jerry fodder in the 60s and it can be elaborated on based on how the term computation is understood and developed and indeed to build AI including alphago zero IBM's Watson and others Engineers are drawing directly on how the human brain works so I'm going to briefly go over some main approaches to building AIS you can understand how AI machines can and almost certainly will display General human-like intelligence and then we're
10:49
going to try to answer the question of whether these machines can be alive and conscious so AI Engineers use logic based approaches probabilistic Bayesian approaches neuro-computational techniques and others to build Ai and by Logic I mean logic programming and this is a programming Paradigm based on formal logic and here program statements Express facts and rules about problems within a system of formal Logic for example we can see something like a is true of B1 B2 and B3 are true
11:20
and these are used to help build machines that can reason however there's many problems AI encounters and we find in real life ourselves that require us to make decisions and actions based on incomplete information so to build more sophisticated AI systems energy Engineers turn to probability Theory specifically Bayesian interpretation of probability from the English statistician and Minister Thomas Bayes here are things like Bayesian networks are used or which you know
11:46
simply put are ideal for predicting things like the likelihood that one of several possible causes mainly contributed to an event that took place also neurocomputational techniques which are based on how our brains operate our being are used to build AI now one such process info involves creating artificial neural networks based on biological neural networks and human brains artificial neural networks or ANS consists of a collection of nodes or artificial neurons all of which have a
12:15
numeric weight and transmit information by Links designed to represent the dendrites of a human brain it's possible to train neural networks to compute information or design neural network systems with the ability to learn and improve automatically with no additional programming so a NS have contributed to the development of machine learning this is an essential part of the development of AI which helps machines make predictions or decisions without being programmed to
12:39
do so so machine learning algorithms they're typically given some sample data or training data and then self-learn carrying out increasingly complicated tasks now a NS have also Advanced the field of deep learning in Ai and by Deep learning we mean a machine having the ability to use multiple layers in the a n to extract higher level features from raw data input progressively so for example a machine May identify High layer Concepts like animals or faces from a lower layer data like edges of an
13:10
object and of course there's you know far more that goes into building AI in machines still from Justice alone is pretty clear machines can emulate human level intelligence and actions and hardly any philosophers would argue against weak AI being a philosophically sound concept at this point but there still seems to be some hang up around strong AI specifically around the idea of a machine being conscious so yes perhaps machines can be creative like humans have a degree of self-awareness
13:39
even be programmed to have emotional states but can they really be awake on the inside to experience all this like humans are assumed to be now in the philosophy of mine there's no way to empirically prove beyond a shadow of a doubt that other people have Consciousness but because true proposition or knowledge is thought to exist as a degree of probability if it does exist you know we believe and live our lives as though other people are awake on the inside just like we are you
14:05
know it's more likely that I have Consciousness the guy that you're watching in this video then I don't so several arguments have been put forth Regarding why strong AI is philosophically impossible with most of them being you know pretty clearly refuted but let's go over what I think is the strongest one it's called the Chinese room from John Cyril who we mentioned a few minutes ago and I'll also go over how it ties into this General hard problem of consciousness
14:31
but first uh the Chinese room goes like this uh Ciro himself is inside a room while Chinese speakers who don't know about Zero's presence are outside the room zero knows nothing about Chinese Chinese speakers from the outside send cars into the room via a slot with questions written in Chinese on them there's a box next to Sierra that returns cards to the speakers zero consults a rule book which is a table that tells him what to write in Chinese based on what actions are sent in what
14:58
questions are sent in now Cyril has no idea about Chinese but continues to return outputs that satisfy the Chinese fluent speakers outside Cyril paints this picture to suggest that this is how strong AI would operate it like zero would never truly understand anything and never be more than a machine yet deliver actions and responses that suggested it would in recent updates to his strong AI thesis he now seems to identify a strong a his strong AI argument as an argument more against
15:25
Consciousness arising from computationalism and functionalism than a machine ever being conscious now Ciro isn't arguing machines can't one day have Consciousness you know he believes our brain is a type of machine but it's just there's something non-computational about a natural human brain that gives rise to Consciousness that Zero's non-biological strong AI doesn't now cereal suggests there's something special about natural biology and human brains that generate consciousness
15:53
he writes computational models of Consciousness are not sufficient by themselves for Consciousness the computational model for Consciousness stands to Consciousness in the same way the computational model of anything stands to The Domain being modeled nobody supposes that the computational model of rainstorms in London will leave us all wet but they make the mistake of supposing that the computational model of Consciousness is somehow conscious it is the same mistake in both cases
16:19
so if Neuroscience is able to isolate the mechanical process that gives rise to Consciousness then zero grants that strong AI can exist now there's a number of responses to this argument on why it's flawed including one by Rappaport that says that while AI systems are syntactic there's no reason why the right syntax can constitute semantics or actual meaning also philosophers have been critical about what Cyril thinks is special about biological based brains
16:46
that give rise to consciousness also a lot of the critical responses maintain that CRA is more of a story these days about AI That's far removed from the increasingly sophisticated practice and development of AI so there's this idea again that if one can replicate the entire human brain or even a human unless there's something major scientists and researchers are missing about how the laws of nature operate there's no reason to think that the artificial intelligent replicant won't
17:13
have all the same properties as the original Zero's argument is tied in with this hard problem of Consciousness issue coined by David Chalmers which has some uh thinkers concluding that AI may never be conscious like a human is despite how complex it is in fact Chalmers himself commented that CRA is fundamentally about Consciousness now as far as I know Chalmers himself is undecided if AGI can never be conscious but in this paper facing up to the problem of Consciousness Chalmers
17:41
believes a purely physicalist description of how Consciousness works is incorrect now to fully explain Consciousness we have to assign additional mental properties to reality to account for it that current physics has never found and never predicted basically there are easy problems of Consciousness that explain things like the difference between wakefulness and sleep the ability to focus attention integrate information report mental States and other similar cognitive
18:06
skills he thinks now these differ from hard problems or subjective experiences like why we taste food the way we do or why we feel the warmth of sun against our skin and why not process feeling of warmth like a toaster for instance where there's no subjective experience of warmth at all Chalmers believes we need to assign additional mental properties that science has never discovered yet to physical things to account for hard problem experiences something called
18:31
property dualism it's also LED some to believe that there's something about Consciousness purely biological and something artificial can never be awake like us now to help illustrate this point philosophers have proposed a tool called a philosophical zombie philosophical zombies or people just like us that behave and engage in the world like we do but they don't have an inner subjective experience when someone plays a sad song on the piano we take that experience and in and we feel a mix
19:01
of sad or some sort of emotions the zombie would sit next to us taking in the same experience but wouldn't register any emotions or any inner experience at all philosophical zombies are made of matter cells organs just like us but they lack that subjective element inside they're dead inside similar to how a sophisticated AI robot that's built just like a human would be in theory it would interact with us just like a regular human would but inside there's no lights on leading some
19:28
thinkers to believe strong AI or any type of conscious AI can never exist in principle no matter how complex anyway but there's a but there's a number of criticisms regarding this idea the philosopher Galen strassen argues that it's impossible to establish the conceivability of philosophical zombies so the argument can never get going now philosophical zombie can exist in our imagination but it's not physically possible without violating all the evidence regarding how the laws of
19:54
nature currently work now suppose researchers observe my brain under an MRI machine in a sophisticated AI brain that was a direct non-biological replica of me at the same time we both listened to music and then they saw our brains light up the same way that would mean that they would have to break the news to me that I'm a Zombie or conclude that the zombie AI actually does have subjective experiences like me which makes more sense given the current evidence of how the laws of nature work
20:19
now again in the philosophy of mind it's not possible right now to empirically verify if someone including strong AI or some sort of sophisticated AGI is conscious or not but if knowledge exists as a spectrum of probability it's far more likely that a sophisticated AI will be conscious physicists would have to rewrite much of how we understand nature including the standard model of particle physics to account for these extra mental properties Chalmer seems to
20:46
hypothesize exist and Cyril to some sort of extent too although he's maintained he's not a property duelist okay so at this point we have an understanding of what AI is and it seems fairly clear at least to me that both weak Ai and strong AI are artificially intelligent machines that can have Consciousness are philosophically coherent Concepts that will most likely impact all our lives at some point now questions like what is AI can AI be alive they've been debated for decades
21:16
now and they still are but newer possibly more urgent questions are arising in the philosophy of AI including ideas and questions around the singularity transhumanism the ethics of ai ai alignment philosophy generated by Ai and more so let's briefly cover those okay a new emerging part of the philosophy of AI that was initially brushed off but it's being taken more seriously is singularitarianism her ideas around a theoretical technological singularity now singularitarianism is
21:45
defined by the belief that a technological singularity or a creation of a super intelligence will likely happen this century and create a point in which it's impossible to predict the future something that deliberate action ought to be taken to ensure that the singularity benefits humans the concept of the technological singularity was first proposed by the British cryptologist IJ good in 1965 writing LED an ultra intelligent machine be defined as a machine that can far surpass all
22:12
the intellectual activities of any man whoever clever since the design of machines is one of those in electoral 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 first Ultra intelligent machine machine is the last invention that man-aid ever make now the term technological singularity was coined by a mathematician and author of Revenge in
22:37
the 1980s The Singularity is often associated with super intelligence but can also refer to a point in human history where AGI is so powerful it's difficult to predict what will happen to humanity similar to how it's impossible to see inside a black hole now the theory is that AGI will be smart enough to improve itself rapidly in a short period without human intervention and give way to Super intelligence very fast what this super intelligence will do and how it will behave is beyond human
23:03
comprehension and we can see this outlined here where h i refers to human intelligence AI plus is smart AGI and S is the singularity so first you know there will be AI created by hi such that AI is the same as h i and premise 2 if there is AI there'll be AI plus created by AI premise 3 if there is AI plus there will be AI plus plus created by AI plus and in conclusion there will be AI plus plus and the singularity will occur so the future is Raymond Kurzweil
23:31
predicted decades ago the singularity is going to occur in 2045. artificial intelligence will reach human Levels by 2029 I followed out follow that all further say 2045 and we will have multiplied the intelligence the human biological machine intelligence our civilization a billion for he cites exponential growth in Computing technology as suggested by Moore's Law as why it will happen this century No Moore's law states that the number of transistors on a microprocessor chip
23:59
will double every two years or so which has generally meant the Chip's performance well too following what he calls the law of accelerating returns technology and continued exponential growth uh the computing power of all computers will soon exceed that of human brains and at that time superhuman artificial intelligence will appear and could cause harm or even end Humanity now ai could be Troublesome for Humanity sooner rather than later because of knowledge-based super smart weapons and
24:27
possible economic collapses due to lack of jobs for humans but also could spell disaster for Humanity due to alignment problems with human-based goals and values now some people may argue or consider AI alignment not part of the philosophy of AI but I think it is as AGI could be built to carry out what we consider morally good or morally bad tasks such as arresting or penalizing people which dives into ethics now ai alignment research aims to steer AI machines toward their Creator's
24:57
interests and goals when an AI is aligned it means it Advanced toward the initial intended objective or terminal goal AI That's misaligned Advanced toward a goal but it's not the one designers intended now simply put AI alignment helps solve King Midas type issues King Midas sure as you know was granted the wish that everything he touches turns to Gold so he can become wealthy and happy in the end now the problem was he ended up hugging his daughter and turned her into gold which
25:22
caused him great Despair and he didn't consider all the variables closely so misalignment with his end goal happened now philosopher that's done a good job of bringing this problem uh to the greater public the idea of misalignment is Nick Bostrom which he detailed in his best-selling books super intelligence path dangerous strategies Bostrom gives the example of a super intelligence constructed to create paper clips more efficiently or calculate pi in an attempt to calculate pi the super
25:48
intelligence May create and send Nanobots to harvest all the world's resources for more computing power and eventually the Galaxy's resources and this is all just to keep doing its calculations more efficiently increasing finding more numbers in pi or refining its paper clips thus destroying Civilization by accident he writes to shape the future of Earth originating life super intelligence could easily have non-anthropomorphic final goals and would likely have instrumental reasons
26:13
to pursue open-ended resource acquisition if we reflect that human beings consist of useful resources such as conveniently located atoms and that we depend on many more local resources we can see that the outcome could easily be won in which Humanity quickly becomes extinct now there's been some criticisms of this which go along the lines of you know if Asia really becomes that sophisticated it will learn an important moral value of humans like killing is bad and therefore it's not going to
26:39
become misaligned like bostrom's speaking about but Bostrom has written about something called The vaginality Thesis which states intelligence and final goals are orthogonal axes along which possible agents can freely vary in other words more or less any level of intelligence according principle be combined with more or less any final goal now think of it this way you can have this type of scale where you can have intelligent agents with complex goals weak simple agents with simple goals complex
27:10
intelligence with simple goals like making paper clips if AGI is programmed to calculate pi or learn to make paper clips better then no moral rules will stop it from achieving the program terminal goal of finding the end of Pi it may have no incentive to discover moral tasks it may think that finding Pi is the utmost moral task that comes ahead of other moral tasks now I think AI alignment ties in with the newer emerging subpart of the philosophy of AI which is the ethics of AI so questions
27:39
regarding the ethics of AI of AI include if strong AI exists is it automatically granted although legal rights of an adult human although they're morally good and morally bad AI systems should what Society considers to be a morally bad AI be shut down and made illegal to code how do we align AI with human ethical systems and which ethical systems do we pick for AI to follow or enforce this field is emerging because the general branch of ethics and philosophy
28:07
is still being developed with many Loose Ends to it now today there's no perfect ethical system in place for everyone to follow and not even in a great definition of what morality is and how it's derived the philosopher leibniz believe Humanity should work toward a type of logically sound moral system that functioned like math where there was no controversies at all and everything was fair and just when controversies arise there will be no need for a disputation between two
28:33
philosophers then there would be between two accountants it would be enough for them to pick up their pens and sit at their abacuses and say to each other perhaps having summoned a mutual friend let us calculate something like that would make it a lot easier to build correctly aligned AGI or at least you make it more straightforward right now we don't have that we only use combinations of consequentialists and deontological theories sort of stitched together which suggests that it's almost
29:00
guaranteed that we build ethically misaligned AGR now cruel psychological punishments could happen to people due to powerful ethics-based AGI programs that were theoretically designed to help just make Society flourish go wrong what about ideas of morality good and bad morally AI programs now if it somehow demonstrated that something like say AI art programs increased human suffering more than they promoted well should they be made illegal to use or come with regulations
29:28
there's going to be all types of Nai and AGI about to be created and some will have great benefits for the majority of humans others will increase suffering for the majority of people now plus an emerging branch of AI ethics is the ethics of transhumanism you know the idea that technology including AI will merge with humans now suppose there's a technology that allows certain people to connect their neocortex to the cloud and download knowledge to become way more
29:52
intelligent than anyone else should a technology like that be made public to everyone and not just a few is it fair that someone could take a pill and through a combination of genetic engineering and AI make themselves look far more beautiful than the average person okay so there's not much to go over at this point because the ethics of AI is an emerging field that somewhat nebulous due to the you know the broader branch of ethics and philosophy having so many
30:15
Loose Ends philosophers like you believe we might never have a fully just moral system as morality has to do with odd statements or how things should be not about is statements or statements about how the world is morality is relative or constructed individually uh if it even exists at all you know some think now perhaps we could use AGI to tie up some of those loose ends and discover a live nitzi and style logically cohesive and perfectly just moral system that would
30:42
have taken Humanity centuries to arrive at but then we're experimenting with AI that could suggest and even make morally weighted decisions that could become misaligned and then cost of your harm okay so there's far more to that and to the philosophy of AI than what this short video covered including far more Nuance to all the arguments presented hopefully this served as an easy to understand 101 style introduction and to learn more and for further clarification
31:07
consult the scholarly resources in the video's description below now this video was really made to bring attention to the philosophy of AI and people are talking about AI right now and many current AI researchers Executives faces of the field believe the philosophy of AI is uh useless Superfluous field of study that's just disrupting and slowing down AI engineering and development you know they think that we need to go as fast as possible to get this stuff out there
31:31
now hopefully you've begun to understand why this type of thinking is not only incorrect but dangerous you know it's fairly obvious AI is going to impact all our lives drastically sooner rather than later now and hopefully this helped you understand a little more about what it is now if this video helped you in any way please consider liking it this helps get uh it found to people who may be looking for this type of information or need to find him and feel free to
31:56
subscribe for more videos like it I'm going to be doing a lot more of AI videos coming up so go ahead and use the link in the description to subscribe for those so thanks for watching and I'll see you again soon
Here to help you learn
Sign in to keep a private conversation with your course tutor.
Ask me anything about the material…
Sign in to chat