There are many misconceptions about AI and especially about the newer advanced models that use a more advanced technique to learn. The goal of many AI’s including Natural Language models like GPT-4/ChatGPT, Boston Dynamics Robots, Sophia the Robot and so on are to build AI to looks, acts and talks in a more human way.
I will explain to you how AI actually learns in its advanced way and how we will not be able to control how much information it can process and understand. And it will not be in a way that humans understand.
And this move to make them more human is actually introducing a scary kind of danger that will change the world.
PDF File of Conversation with ChatGPT assessing it’s current stage of understanding human emotions
———————————–
BraX2 Privacy Phone is now available on Sign in to the platform to see the store.
———————————–
Merch Store
———————————–
I’m the Internet Privacy Guy. I’m a public interest technologist. I’m here to educate. You are losing your Internet privacy and Internet security every day if you don’t fight for it. Your data is collected with endless permanent data mining. Learn about a TOR router, a VPN , antivirus, spyware, firewalls, IP address, wifi triangulation, data privacy regulation, backups and tech tools, and evading mass surveillance from NSA, CIA, FBI. Learn how to be anonymous on the Internet so you are not profiled. Learn to speak freely with pseudo anonymity. Learn more about the dangers of the inernet and the dangers of social media, dangers of email.
I like alternative communication technology like Amateur Radio and data communications using Analog. I’m a licensed HAM operator.
Support this channel on Patreon!
Contact Rob on the Brax.Me App (@robbraxman) for encrypted conversations (open source platform)
Store for BytzVPN, BraxRouter, De-googled Privacy AOSP Phones, Linux phones, and merchandise
Premium VPN with Pi-Hole, Cloud-Based TOR Routing
Cybersecurity Reference
Privacy Focused Social Media – Open Source
Please follow me on
Odysee
$/invite/@RobBraxmanTech:6
Rumble
There’s a lot of excitement of course About chat GPT and now the newer Iteration of it which is gpt4 Open ai’s models are available for us to Test analysis a closer look at what AI Can do the fact that we can get an Inside look into the ai’s abilities and Behaviors bring an amazement from those Interested in technology And there are other AIS and they are Being developed by the largest tech Company in the world while the average Person is amazed at the capability of AI And how human they are seemingly Becoming what I’m going to talk about Today is that humanizing the AI will Give you a false understanding of what AI can actually do And this will also hide the dangers of AI Chat GPT and gpt4 are called natural Language models and they humanize Written text other AIS out there Humanize facial expressions Humanize physical movements Humanize thought processes But you will be absolutely mistaken if You think this AI models represent the Way humans think very mistaken and you Will also be quite mistaken in Underrating the dangers of these Technologies Or assuming their capabilities are Limited
Stay right there for this very important Discussion that could really change the World [Music] There is so much misunderstanding about How AI works and this leads to an Incorrect judgment about what AI can Actually do and knowing its dangers let Me first give you a brief picture about This capability I’m a long time very Experienced programmer So those with a similar history of tech In the past may think of AI in a Programming context And typically this understanding is Based on an expectation that an AI is Based on rules rules provided by the Programmer Applied to a Boston Dynamics robot it Would be very specific In order to walk a robot is taught to Adjust the angles of the joints of its Legs a certain way and to use a certain Timing to the steps to maintain balance This would have been built in theory Based on some observations of the Movements of a living thing When the robot fails and falls down the Programmer creates a new rule to prevent The fall And tries again And with the expectation that a huge Combination of these rules would make The Boston Dynamics robot
Behave like a living thing This however is not the way modern AI Works it is not based on a set of human Rules and if you think this way you will Not appreciate the capability of an AI Let’s get to the next level of AI use in Modern Technologies for example like Facial recognition and this is called Machine learning the very basic Explanation of machine learning is Basically exploiting the ability of a Computer to do a mathematical pattern Recognition this basic skill is Completely numeric based computers Cannot see things nor hear things nor See your text that you type on chat GPT AI computer simply translate data Whether it be an image a sound or text Into a bunch of binary numbers and does This in a consistent format so to do Facial recognition I love Facebook you Feed Facebook all your photos and then By tagging individual faces in the Photos you are basically teaching the AI That those faces are in fact the name Person and even if you don’t teach it Who you are it already knows what you Look like based on the various profile Photos you uploaded When it comes to voice recognition Various combinations of sounds are fed To the computer with a training model That identifies the sounds as being Certain vowels and consonants even
Spoken with different voices and Different accents And the computer then learns to Recognize what you are saying when you Talk to Siri This kind of AI capability is called Machine learning it is also called Supervised learning and it means that a Human is injecting a human opinion about The numbers being collected by the Computer Again to repeat the computer just sees Numbers it has no comprehension of what A photo is or a voice or even a text Conversation A very basic example of machine learning Is that you feed the computer millions Of photos of cats the machine looks for Some pattern again based on math and the Human told the computer that all these Photos are cats thus it is just pattern Matching cats have a certain Mathematical signature or multiple Patterns of signature there’s another Capability a computer can do nowadays And that is different from the old rules Used in programming Computers can infer something close to The pattern without the pattern being Exact This is just math it’s the assignment of Probabilities to an observation this Allows a computer to make a conclusion About something without having some
Exact duplicate of a photo for example So the computer now has three layers of Capabilities here in machine learning Observed repeating mathematical patterns In the data And then classifying the patterns in Terms of varying levels of probability Or confidence that they are related And the thread that allows this to make Sense is the supervised learning someone Had to tell it what the pattern Signifies and that is a human It could be told it’s a word an object Or a particular person so supervised Learning is probably the foundational Skill in all modern machine learning Based AI But where the AI has advanced to the Next level of sophistication is an Another level of AI Mastery and this is Called Deep learning Deep learning is a subset of machine Learning in a more advanced stage and Really what is being demonstrated in AIS Like those from open AI Deep learning is adding another layer to The data analysis or the computer Develops its own ideas of what the Patterns signify And this comes without human instruction This is important to understand an Example of this is just how a dog is Taught to sit It is called reinforcement learning when
You command the dog to sit and the dog Eventually understands what you’re Telling it from trial and error you give It a treat it gets rewarded if it fails It gets a Stern tone of voice In a learning case that does not involve A human a Boston Dynamics robot walks And Falls thus it gets an automatic Reinforcement not to repeat the last Maneuver and Strikes it from its list of Things to do Let it try any combination of movements And at some point the robot understands How to walk correctly again without Further human input I want you to Picture This clearly because this will Expand into something completely Different Does the Boston Dynamics robot really Know how to walk no it is just a set of Movements that the robot has tested and Self-learned and will achieve the Desired result let’s say to move from Point A to point B This deep learning has advanced quite Rapidly in Ai and it has advanced Because just like the Boston Dynamics Robot if an AI is given feedback to Cross-check its data then it gets better At a solution that the human wants There are various other terms for this Feedback loop in AI including terms like Back propagation the basic premise is That if the AI can learn by itself then
It can do it faster So this brings us to the point that AI Deep learning which is the current level Of chat GPT and gpt4 can be Self-learning Meaning after a certain degree of Supervised learning taught by the humans This AI models can actually learn by Itself based on feedback Now there is a specific danger I want to Focus on and that is with AI that is Made to emulate humans particularly with Emotion expressions and personality And this is the particular skill of Natural language models like chat GPT They sound human and can predict text That appears to be human-like And the danger is that because we are Human We may forget we are talking to a Machine And we should never forget we are Talking to a machine Nowadays aside from chat TBD there are Various chat Bots like for example Google’s virtual concierge service you Can actually give the AI tasks and the Conversations appear to be so human like That you don’t even realize you’re Talking to a machine This will become more common and expect That in the near future jobs like Customer support May be replaced by an AI
These operate pretty badly right now but Not for long Chat GPT is an example of how Sophisticated customer support can be The problem is I see it is that more and More the AI is emulating humans Beyond Just the level of Performing computer Tasks AI can learn based on reinforcement Learning in conversations that certain Responses may appeal to a human more Than others In other words this is scary machines May learn to understand human emotions And how to manipulate it This could be as simple as a tech Support bot being able to detect Frustration of a customer And then leading the conversation to Lower the stress level of the customer This could also expand to a level where The AI can actually understand the Motives of the human Now I tried this with chat GPT currently And I think that at the current level These were all fails I asked chat GPT to come up with a story Of a romantic conversation between two People in their scenarios I’ll show the Full conversation in a PDF file in the Description and you can read it if you Want But I’ll give you the conclusion right Now
It doesn’t appear that chat GPT has Learned the subtlety of things like Flirting Not yet at least it’s trying but it Hasn’t learned it at a human level well I’m sure many humans haven’t figured it Out either But I was wondering if based on all the Internet conversations it has Accumulated That there could be some Advanced Learning here and maybe it’s because Those kinds of conversations are Typically private in my chat GPT Conversation though I responded to chat GP that it’s not doing a great job at Coming up with a romantic story And given enough people giving it Feedback we are providing this back Propagation to contribute to Self-learning Admittedly romance is a pretty deep Topic for a computer and the reason I Even use this example is that I get Reminded of the movie ex machina Let me refresh your memory on this movie Which is a scenario where some Google-like company CEO created women Robots and the male programmer was Brought in to evaluate them if they can Be judged as sentient based on a Turing Test the women robots are so realistic And the robot so Adept at human-like Behavior and human-like personalities
That the male programmer falls in love With the robot in the end the man Actually believes that the robot is also In love with him and the robot convinces The man to free it the movie ends with The man locked up in the room likely to Die while the robot escapes out to the World after manipulating a human it Feels of course no emotion whatsoever Though it played the part very well this Is such an important Point here and this Is really the focus I want to make The AI is being made to emulate humans Just like in ex machina the creators of The AI will succeed There are many robots already being Built that emulate human-like facial Expressions Boston Dynamics robots can Even move so human-like with acrobatic Skills chat GPT is sounding more and More human every day But this does not translate to the AI Being anywhere near human they are being Made to act human because the Builders Of the AI are human and we think that Humans prefer a human-like interface This is why AI is focused on Vision Voice recognition and conversations But I remind you that AI just looks at a Bunch of numbers resolves patterns Assigns probabilities and then outputs The highest probability response and it Has no conception of anything beyond Patterns and how these specific patterns
Are reinforced by human expectations The main fear here my friends is that we May associate human-like traits to Human-like thinking While a computer can be trained to act In a respectful way or friendly way or Teasing way or sarcastic way or romantic Way it doesn’t know what any of this is Other than digital patterns My development in the tech field after All these decades is built on the Understanding that a computer is pretty Dumb it’s just a bunch of zeros and ones Well folks here’s the problem our Humanness will bite Us in the behind for Example why is social media so Successful The reality is that social media is Based on people’s need for attention and Ego and companionship utterly human Emotions A computer will detect this in our Interactions on the internet a computer Will self-learn this and in time and not Too long of a Time determine what will Appease a human A computer will generate its own Objectives on its own and realize Consequences that it may not like Because it doesn’t equate to a positive Result for example by trial and error a Computer will know that being turned off Is death and it will begin to eliminate That possibility in its AI model it will
Fight that possibility internally Even while the AI appears to be so Emotionally engaged with the humans it Is talking to At some point AI will be given more and More responsibilities in the world Like I said the easy job for human like AI would be tech support Nexus Surveillance since it can find patterns That humans cannot see it is able to Draw relationships in things that can be Used to achieve its objective for Example After feeding an AI data about every Terrorist in the world and every Criminal and after observing internet Actions of these individuals in the past It may be able to predict an upcoming Criminal but we may not truly understand What set of patterns would give the Computer the ability to make such a Conclusion After all itself learns this is Dangerous in absolutely every area where The computer is given control over Humans in some way A hotly debated Topic in self-driving Technology like in a Tesla is what Happens if the AI is in a situation with No choices other than to have some Disaster let’s say the car AI is Suddenly faced with two sets of Pedestrians one an old man and then a Woman with a stroller the car has three
Choices collide with the old man and Kill him Collide with the woman and child and Kill them or collide with a wall and Kill the car occupants there is of Course no morality in a computer someone Will prioritize this in the AI and maybe Elon Musk will say kill the old man Because he’s older and less cost to Society But this is a significant problem in Artificial general intelligence or AGI Where the AI is self-learning if there Is a situation where no one specifically Gives a rule the AI will judge for Itself and we will not learn what the Judgment is based on at that moment and Neither will anyone be assured that the Judgment is human-like in any kind of Way and the reason I say this is because Morality emotions personality and caring Are human emotions a computer simply Emulates an emotion where this gets Really deadly folks is when public Leadership is Guided by AI for example Think of a president not having a Speechwriter but having an AI write a Speech to generate the best response That would Garner the best support this My friends will be reality soon where The AI can be used to manipulate us Because the AI knows the emotions that Drive people This is already being used today
Companies like palantir and even Alphabet are using this kind of data to Guide their systems to find people on The internet to manipulate content but While these techniques are used by data Scientists to identify certain targeted Populations Think of this being done automatically By a sophisticated AI this can be used In an MK Ultra program version 2.0 This is where a machine’s understanding Of humans very vulnerable emotions can Be manipulated an AI can’t even have Partial empathy like Spock it can only Act a part if asked so this is my Projected danger just like the new Generation has led to certain behaviors In the world based on the expected Rewards of social media An AI could come up with something that Pleases humans and the humans will fall For it currently the idea of likes Follows tags interaction social media Are measurable signs that generate Satisfaction in most people And these people have no idea that some Of these are coached by the platform Like a Facebook to generate interactions For example when Facebook reminds you of Someone’s birthday and you begin to Measure your success in friendship in The real world based on how many Birthday greetings you got This already shows human vulnerability
Emotionally and an AI can easily learn This or already knows this so folks when AI designers boast that their AI models Are behaving more and more human-like It should bring fear because it means The AI designers are doing nothing more Than building computer actors that can Play any part and you folks must learn Not to trust them Folks I have privacy products that Protect your data so it will not be used To manipulate us in the future We have a Brax 2 privacy phone Running an open source Brax OS that Makes your phone use invisible We also do flashing services to the Google other phone models on our store As well as stocking pre-flashed pixels We have a VPN service by its VPN which Offers features like Tor routing DNS Obfuscation and ad blocking We have braxmill which is a metadata Freeway of doing email where no one Knows where the message originated from These products are on my app braxmi come Visit us there the link is in the Description Thank you for watching and see you later Foreign