Learn about the latest AI tools available in business, like AI video generators, audio, video game creation, and autonomous vehicles, a brief summary of how AI works, the legal ramifications of AI, and what jobs could be up in the air by 2030, 2040, and even 2050.

Explainer

AI in 2023 and beyond

Synopsis

What do these three things have in common? A game of Pacman, recreated from scratch, Joe Rogan interviewing Steve Jobs in 2022, and the prediction of melanoma and cardiac dysfunction risk in patients? If you guessed artificial intelligence, you are correct!

In this report, we share the latest AI tools available in business, like AI video generators, audio and video game creation, predictive models, and autonomous vehicles; a brief summary of how AI works; the legal ramifications of AI; and what jobs could be up in the air by 2030, 2040, and even 2050.

The top examples of AI

Get inspired and terrified: here's a taste of what AI can do going into 2023:

  • AI can now believably impersonate humans on phone calls to schedule appointments.
  • Self-driving taxis are available in downtown Phoenix. Ikea and Walmart are trialing driverless truck deliveries in Texas.
  • Autonomous black hawk helicopters can now fly one hundred and thirty four kilometers.
  • Google's Wing already delivers for DoorDash customers in Australia.
  • Apps like Lobe can train your own machine-learning models to identify plants, recognize gestures and emotions, or count reps.
  • AI music has existed since Alan Turing's computer-generated melodies in 1951. Today, there's a host of platforms that can generate AI music. There's even an AI Song content for humans that use AI to make music.
  • AI can generate realistic images from text prompts to imitate famous artists, win the Colorado state fair, and create the art for entire video games and comic books.
  • Apps like SceneryAI can isolate specific parts of an uploaded image and replace them with AI generated components.
  • Another tool, Unscreen, uses AI to remove the background from videos.
  • Tools like Facetune can edit, enhance, and retouch selfies. AI can even decide how attractive you are.
  • Google's Imagen Video can create short clips for phrases like "a teddy bear doing dishes."
  • And Google's DreamFusion system can create 3D assets from prompts.
  • In 2020, AI made by NVIDIA was able to recreate a playable version of the game Pacman based on simply watching hours of the game being played.
  • AI has also been used to design game levels, with one AI system used as a player to test the levels while the game AI builds them.
  • Game developers already use tools to create compelling professional voice performances.
  • The tool Altered AI features 20 professional "AI actors", and has been used by indie games and even a Triple-A game studio.
  • AI is now used to write articles, screenplays, and fantasy novels.
  • Creators use Otter AI to dictate a video someone else created, and then use Jarvis or Jasper AI to change the text to create their own version.
  • OpenAI's GPT-3 large langue model can already fool human readers, so Google is actively trying to stop AI-generated articles.
  • Google's own attempt to develop a "universal speech model" trained on 400 languages, could be the largest leap forward yet. With hundreds of languages in one AI model, Google could improve its search engine across multiple languages.
  • AI is used to analyze penalty kicks and jump shots, like with the HomeCourt app.
  • AI can also be used to track the movement of every player to make an AI version of the entire team for training.
  • And Google can even create robots that autonomously generate new code to meet human demands.

And that's not even 1% of what's currently available or on the horizon!

It's been predicted by MIT that the next breakthrough in AI will come from multimodal AI models that can use computer vision and audio together to interpret information. Large language models built into AI as they sense the world could help robots understand their surroundings through visual and vocal cues. Combined with reinforcement learning, AI will soon explore with autonomy and interact with their environment to learn even more.

How AI works

How AI works is beyond the scope of this video, but here's a high-level crash course. To count as AI, a system needs to make its own decisions, and make its own predictions. Developers use machine learning to generate "artificial" intelligence, or thousands and thousands of decision-making units that are interconnected in a lattice that all work together to learn patterns. These lattices form a neural network, an information processing machine meant to simulate a brain, with interconnected networks of neurons that make different decisions and predictions based on different inputs.

The AI's artificial neurons take in inputs and create an output of either 0 or 1, with unique weights and biases that can be toggled up or down depending on what's being trained. These numbers, stored in each artificial neuron, form the memory of the AI brain. All this computation and learning requires a ton of input (as in datasets of hundreds of millions to even billions of inputs) so that the AI can learn.

The way the AI teaches itself how to weight each input is called backpropagation. Developers give the AI training examples or inputs where the desired output is known. It creates predictions from this, and an error score is assigned to each output. The machine then rebalances itself backwards over time to learn the optimal weights to minimize errors and make more and more accurate predictions. And that's generally how AI works.

Legal ramifications

So who owns the copyright behind AI? AI has been on the US copyright office's mind since 1965 when someone tried to copyright a musical composition made by a computer. The US copyright office ruled at least twice that AI art can't be copyrighted unless there is a directly responsible human collaborator. The UK, EU, and Australian intellectual property offices have ruled the same. There's actually a four-step test to determine if AI can gain copyright in the EU:

  1. It has to be a production in the literary, scientific, or artistic domain [item]It has to be the product of human intellectual effort
  2. It has to be the result of creative choices
  3. And those choices are then expressed in the output.

The third step there is the most important: if a work does not reflect the free and creative choices of the author, that work is considered in the public domain, copyright speaking. Until determined otherwise in the court, humans who use AI as a tool should be protected by copyright as the owner. But AI tools where a human enters a text prompt to create an image, don't currently qualify as an original work of authorship and therefore can't be copyrighted.

So what if an AI recreates an image of you? As of 2020, anyone can make realistic deepfakes without the use of complicated software. And AI already exists that can copy the voice of anyone with voice imitation algorithms. This could lead to an increase in identity theft and fraud. In 2020, a bank manager in Hong Kong fell victim to a voice-deepfake that cloned the voice of a director at another company to authorize $35 million dollars in bank transfers. This was the second known case of AI voice cloning to carry out a heist, but the first successful one.

And then there's the legal gray area of the rights over to your voice and body. The British actor Rick Kiesewetter signed away all of his rights to recordings of his voice and facial movements as part of a job for a tech company. Could this company one day legally recreate its own AI version of Rick and profit off of it without paying him?

And then there's not safe for work applications, such as deepfake sexual content, which can be used to create images without the subject's consent. In 2019, the research company Sensitivity AI found that 96% of deepfake videos online were non-consensual adult content. The open-source image generator StabilityAI has already been used to recreate Not safe for work content of celebrities. So if an actor like Rick signed away the rights to his voice and movements, could it one day be used to legally recreate adult content of him without his knowledge or consent? Parallel movements in the US and UK are gaining momentum to ban nonsensual deepfakes. For all these issues, a wave of litigation is on the horizon to hopefully provide us answers over the next ten to fifteen years… when there's enough money on the line, at least.

AI's impact on jobs

A 2022 survey of 2,000 workers found 14% reported losing their jobs to a robot. The futurist Thomas Frey estimated 50% of today's jobs will no longer exist by 2030. By 2024, AI is predicted to be better at translating languages than humans. By 2026, they could write better high school essays. By 2027, they'll be better at driving trucks. By 2031, they'll work in retail. By 2049, they'll write a bestselling book. And by 2053, they'll perform surgery. Its also been predicted that all human jobs could be automated over the next 120 years.

Here's a brief summary of the industries (and roles) that will be impacted:

  • Taxi drivers and cashiers are most likely to be replaced by machines
  • Deliveries will be mostly by drones or self-driving cars
  • Telemarketing will likely be fully automated
  • Other roles like accountants, construction workers, lawyers and doctors won't be fully automated, but will be artificially augmented.
  • For accounting, work that used to be done by humans is already performed by computers.
  • In law, Deloitte predicts that 100,000 legal roles (grad students who comb through research) will be automated by 2036.
  • IBM's Watson AI already diagnoses lung cancer at a 90% accuracy rate compared to 50% for human doctors.
  • Pharma companies use AI to cut the time to investigate uses of new drugs from one year to as little as two weeks.
  • In aviation, robots administer a coat of paint on an airplane wings in 24 minutes, which would take the humans hours to do.
  • Amazon unveiled a robotic arm that can identify, lift and sort approximately 65% of its product inventory. 75% of Amazon's 5 billion packages processed annually are already handled by robots at some point.

The scale that AI provides is the issue, as it encourages the centralization of wealth. The few major companies with the best algorithms will likely own the market, creating a winner takes all environment. For example, the safest autonomous taxi company will be the one everyone rides. The best image generator will become the one everyone uses instead of contract artists. Adobe Shutterstock will now license and sell images exclusively created on the Dall-E 2 image generator. While the plan is to compensate artists with royalties when their IP is used, individuals are banned from uploading their own AI-generated art to the platform. As these services get more advanced, their costs will rise, and artists will have to undercut their own rates in order to keep up.

There are only two solutions: regulation or greater investment in AI for more competition. You can't afford to be left behind, so if you aren't currently using AI as part of your business, it's time to get involved. When it comes to the future of AI in business, it is truly unlimited - at least for now. Thanks for reading.