Where is AI headed?
Below are set out the current general trends in AI development. This is from the point of view of the public perspective on AI. In addition to the widely accessible AI that has recently burst onto the scene, AI has been used behind the scenes for a number of years. It is also increasingly being used for research in medicine and science. As a result, over the next few years, we will see a series of major breakthroughs in science and medicine that will have been accelerated by the use of AI.
In terms of the general development of AI, as you can see from below, most software, systems and even many objects we interact with will soon have AI embedded into them. In addition, we are rapidly moving towards a world where text, image, video or audio can be quickly and cheaply fabricated. Soon it will be impossible for the untrained or unAI-enabled eye to tell whether any text, image, video or audio is an actual representation of what happened or is totally fabricated.
Robotic systems, which are rapidly becoming more agile and stronger than humans, will soon have AI embedded in them. This will make them able to communicate and problem-solve in a way that they have not yet been able to. This will make them much more attractive for companies to use. As with AI in general, this will have unknown effects on the employment of human workers.
If we all had confidence that governments could move rapidly enough to effectively manage the introduction of AI and ensure that its benefits are shared fairly, we would all be cheering for the wonders that AI can accomplish. However, many people do not feel that governments will find it easy, or easily affordable, to respond to AI in time and in the most socially beneficial way. It is important that individuals and civil society get up to speed on the potential social, psychological, economic and political impacts of AI and then let governments hear their voices on how they think the introduction of AI should be handled.
Current steps in AI’s developmental pathway
Step 1: Teaching AI To Understand and use language - Chatbots
We can start our overview of how AI development is proceeding with the recent release of accessible chatbots (also called large language models). The best-known of these is ChatGPT. Chatbots are AI systems trained on large amounts of text from the internet which can answer questions and produce a range of text products. These products include generally chatting with a user, answering questions, producing academic-sounding reports, writing business plans, and drafting speeches etc. OpenAI has produced ChatGPT and Microsoft has incorporated it into its Bing search engine. Google also has a chatbot called Bard. Other companies are also producing their own chatbots.
Demo: Short video on what ChatGPT can do.
Step 2: Building chatbot intelligence into other software
While it is still just a few months after the introduction of chatbots, we have already seen the next step in the your of AI. This is to incorporate chatbot intelligence and language abilities into other software and platforms. This can be called AI-Enhancement. An example is Microsoft Copilot. Google, and other software companies are rushing to do the same thing. Social media platforms are rapidly moving to allow users to interact with AI agents on their platforms.
Demo: See a short video on Microsoft Copilot.
Step 3: Allowing Chatbots to themselves use other software
A further step is now also being implemented. This is to not only include chatbot intelligence and communication skills within other software, as in Step 2 above. This third step is to now allow chatbots themselves to make use of other software and internet platforms. ChatGPT is doing this by allowing ‘Plug-Ins’. This is similar to the Plug-Ins that we use with some internet browsers. Microsoft Bing search engine also allows a version of ChatGPT to go away and search the internet when you ask it a question. Google is doing something similar with Google search. Another example is if you the ask the paid version of ChatGPT to draw an organisational chart, it will now work out itself that it needs to go away and activate a chart drawing Plug-in. It will then show you the diagram which that other piece of software created. This makes chatbots a lot more powerful and useful. Even more advanced systems are now being developed. For instance, AutoGPT can accept instructions, divide what it needs to do up into steps, remember the order in which to do these steps and access other software to achieve its goal.
Demo: Microsoft is using ChatGPT to provide AI-powered answers in its Bing search engine, see a short video on how Bing works in this mode. Seen a brief video on ChatGPT ordering some groceries. See a short video on AutoGPT.
Step 4: Letting chatbots produce multimedia outputs and receive multi-inputs
The next step, parallel to the development of chatbots, has been to build AI systems that can create images. These are called AI Image Creators or Generators. They allow the user to write a prompt (e.g. ‘Draw me an image of an elephant on a bicycle’), and such systems will produce an image of this. In theory it is now possible, when chatbots are replying to a question from a user, for them to be able to fabricate images to go along with the text that they are producing when they respond. In addition, chatbots are now being built that can allow image imput, not just text input (e.g. ChatGPT has this ability, but it has not yet been turned on for the general public). Soon chatbots will be able to accept a wider range of inputs than humans (text, sounds, images and a wide range of sensors). They will also be able to output a wider range of outputs than humans, including text, sounds, images and instructions for other software and systems.
Demo: See a short video on Bing Image Creator which is a free example of an image creator. Midjourney is an example of a paid image creator.
Step 5: Developing AI that can manipulate images
Another parallel development has been to develop AI that can manipulate existing images. The recently announced GragGAN AI system will let the user take an image and quickly manipulate it in various ways. For instance, you can make a person in a photograph who is currently smiling, stop smiling. This has a wide range of implications for fake news and deep faking.
Demo: Brief video example of changing a picture of Joe Biden from him smiling to him not smiling.
Step 6: Helping AI to mimic people’s voices
At the same time, AI has been developed that can listen to any human voice and then mimic that voice. This means that you can create audio of anyone seeming to say anything you like, regardless of whether they ever said it in reality. Play.HT is an example of this type of AI.
Demo: Listen to some brief synthesised clips mimicking famous people.
Step 7: Getting AI to both create and manipulate video
Up until now, the quality of AI-created and manipulated video has not been up to the same standard as AI-generated text, images and audio. But AI video creation and manipulation are improving fast. This means that videos of anyone saying and doing anything will be easy to generate. It is hard to estimate the exact time when credible video faking will be of the same standard as still image faking, but it is coming.
Demo: See a short video of an example of a politician’s face being substituted for an actor in an episode of The Office.
Step 8: Building powerful robots
At the same time as the above steps are taking place in regard to AI, work on robotics is also advancing rapidly. Agile and powerful robots are being developed and refined. For instance, the robots that are being developed by Boston Dynamics.
Demo: See this video of an example of a strong and agile robot that could be used to assist a builder.
Step 9: Putting chatbots into robots to make them more intelligent and able to communicate with humans
The next step that is now being taken regarding robotics is to put chatbots into robots. This will make robots more intelligent and better able to communicate with humans. In the demo below there is an example of a robot showing expressions on its face. ChatGPT has been put into this android and as a result the interviewer can talk to it and it will respond sensibly to the questions that the interviewer is asking. Notice how, when the robot responds, it shows emotions on its face that are consistent with what it is discussing with the interviewer.
Demo: See a video here of a robot showing facial expressions as it understands what it is discussing.
What will happen next with AI?
AI will make it much easier to use all sorts of software, internet platforms, systems and even objects (such as cars) much easier to use.
There are many implications of this for enhancing human lives. This will particularly be the case if resourcing is provided for communities to work out how they can use AI in creative ways and to deal with some of the major problems of our time such as: climate change, inequality, educational inequality and social exclusion.
AI will lead to major advances in medicine and science.
There are likely to be huge implications from this in terms of new treatments and new technology. If the benefits from these are shared fairly then AI could massively improve health and many other aspects of human life on earth.
AI will rapidly become faster and more intelligent than humans (in regard to some aspects of intelligence).
There will be unknown disruptive social, psychological and job market implications from this.
A flood of fabricated text, images, video and audio will make it much more difficult to know what is true and what is not. This will result in fake news, deep fakes and widespread spamming and scamming.
Unknown political and social consequences will arise from this wave of misinformation.
AI will be built into most software and systems and a number of the things that we interact with.
Many benefits are likely to come from this.
AI will be able to scan and summarise massive amounts of information about individuals.
This will be able to be used for a wide range of positive uses but there are likely to be implications regarding surveillance and privacy resulting from this.
AI-enhanced robots, much stronger, more agile and faster than humans will be developed. Some of these will be military robots and AI-enhanced weapons systems.
Such robots will be extremely useful in many situations but there are also unknown implications for the job market and society at large.
Welcome to the AI revolution!
AI development is moving so fast that it is important that as many people as possible and as many civil society organisations quickly come to understand the potential impacts of AI. Doing this will enable them to turn their minds to what the rapid introduction of highly intelligent AI and widespread AI means for society as a whole and how the introduction of AI should be best managed by government, companies and communities.
Image credits: Woman looking to the future https://www.freepik.com/free-vector/girl-standing-growing-arrows-looking-through-binoculars-woman-searching-opportunity-job-flat-vector-illustration-business-strategy-goal-pathway-career-concept_29119173.htm. Language https://www.freepik.com/free-vector/language-composition-with-flat-design_2609860.htm. Chatbot with other software around it https://www.freepik.com/free-vector/call-center-isometric-concept_26761063.htm. Computer with online booking https://www.freepik.com/free-vector/isometric-icon-with-people-talking-chatbot-booking-website-3d_16717120.htm. Multimedia https://www.freepik.com/free-vector/antigravity-technology-with-elements_4083529.htm. Edit image https://www.freepik.com/free-vector/design-team-concept-illustration_11641475.htm. Voice https://www.freepik.com/free-vector/voice-chat-concept-illustration_19895937.htm. Video Image https://www.freepik.com/free-vector/influencer-recording-new-video_7815291.htm. Robot and AI https://www.freepik.com/free-vector/chat-bot-concept-illustration_13317063.htm. Robot https://www.freepik.com/free-vector/illustration-robot_2606096.htm.
Copyright 2023 Dr Paul Duignan.