eWEEK content and product recommendations are editorially independent. We may make money when you click on links to our partners. Learn More.
In the popular imagination, artificial intelligence (AI) is often thought of as one distinct technology. The reality is that AI is a large concept encompassing a wide range of categories, use cases, and solutions.
Let’s explore the many types of AI.
Also see: What is Artificial Intelligence
What Are the Different Types of AI?
AI category definitions differ widely. Some overviews compare current AI uses with speculation from researchers about what milestones the field might eventually reach in the future.
For the purposes of this article, we will focus on what is currently possible and achievable.
Also see: Top AI Software
Predictive
Predictive AI projects a likely future state of a data point given past actions or some other context. This category encompasses recommendation engines, spell checks, and other applications of AI that use statistical techniques and feature engineering.
This form heralded major advances in AI. Examples include auto-correct on a mobile phone, shopping recommendations, and digital ad spending recommendations.
“The goal here is to accurately predict the ideal next step (and anything anomalous besides it),” said Nagarajan, co-founder and CTO of BigSpring.
Model and detect
Given a large enough dataset and an objective function, users can build a model and run it against new data. Computer vision and general search algorithms fall into this. In addition, model and detect AI is valuable for spam detection, facial recognition, and more.
Generative
Given a prompt, generative AI can create an entire video or code or essay that will pass for human-level accuracy in a second or two. This is the current major wave of AI with tools like ChatGPT, Stable Diffusion, and Github Copilot. They use transformer models that can be generalized to large classes of digital work and can replace chores or whole sets of jobs in a reliable and fast way.
Supervise, unsupervised, and reinforcement learning
Frans Cronje, CEO and founder of Dataprophet, listed the three main subsets of AI as being supervised learning, unsupervised learning, and reinforcement learning. Each can be shallow or deep.
Deep learning, Cronje said, employs a compositional AI algorithm, consisting of a network of layers of nodes known as a neural network. Moreover, deep learning is great at generating meaningful intermediate feature representations from a mass of raw data — provided it is properly contextualized. With sufficient training data, it has worked for pattern recognition tasks like image captioning and machine translation.
What Are Key Use Cases for AI?
AI applications and use cases have evolved enormously in recent years. Here are some of the prominent use cases:
Manufacturing production process control
Cronje cited prescriptive control of production processes in their entirety as an evolving use case. His company has trained deep learning models for preemptive manufacturing process optimization across a range of verticals. This includes iron foundries, automotive, welding, metal working, mineral processing, and semiconductors.
Creating talent pools
Predictive and generative AI can enable companies to create talent pools. Google Ads, for example, used this approach to take the lag out of the deployment of around 500 feature updates a year.
“They suffered a 12- to 24-month go-to-market lag because the team could not deploy skilling to ads practitioners at pace with their product evolution,” said Nagarajan.
Also see: The Future of Artificial Intelligence
Automotive
The automotive industry is very much in the forefront of AI innovation. Yoav Banin, chief product and business development officer at Nauto, highlighted three applications of AI-based systems for the industry:
- Infotainment systems and speech-based navigation based on HMI (human-machine interface), voice command, gesture recognition, and natural language processing (NLP).
- Systems that detect and identify internal risks such as drowsiness detection and seatbelt violations, which is accomplished by driver monitoring systems using cameras and real-time evaluation of the state of the driver.
- Systems that detect and identify external risks such as tailgating based on radar-based detection, sensors, and engine control units.
“We focus on maximizing the utility of AI in helping a human driver perform better rather than trying to completely replace the human,” said Banin. “It has demonstrated superior safety results in terms of lives saved, collisions avoided, and reduced risk due to driver behavior.”
Nauto achieves this through AI-based distraction detection, monitoring in-cabin behavior, and combining computer vision and AI to understand driving context. The system can provide a real-time alert to the driver to enable self-correction.
Where Is AI Heading in the Future?
Natural language processing
NLP is a branch of AI that aids computers in understanding how humans write and speak. These capture meaning from an input of words and produce an output that can vary depending on the application. This is a red-hot area of AI development.
“Most AI trends have to do with natural language processing and how you can use it to make AI models better,” said Ricardo Michel Reyes, Chief Science Officer at Erudit. “As this evolves, businesses will make better sense of all their stored textual data to become data-driven organizations.
“If businesses found themselves collecting more and more raw, organic communications data in past years, 2023 will see a surge of AI tools that help them convert company data into actionable insights.”
Quantum computing
Quantum computing harnesses quantum mechanics to solve problems that are regarded as too complex for the computers designed using the classical system of logic and computation.
AI is being implemented in a quantum setting for more rapid and accurate computation of machine learning algorithms to achieve results that are not possible with traditional computers. It has applications in the processing of large AI datasets, faster solving of complex problems, and integration of multiple datasets.
Creative work
Beyond analyzing historical data to identify the right target audiences, AI use cases are being taken a step further by implementing immediate changes in response to real-time performance indicators.
James Brooks, the CEO and founder of video distribution platform GlassView, said AI is streamlining the creative process at ad agencies. This includes changing aspects of an ad such as the color and positioning in real time as part of ongoing campaigns.
Hiring
Glider AI CEO Satish Kumar said AI is being used in human resources (HR) and hiring, including spotting candidate fraud during the interview process. Fraudulent schemes include acts of copy and paste and code plagiarism.
More elaborate scammers hire professional interviewers at $150 per hour with the candidate lip-syncing answers using earpieces. AI can spot where a candidate is lip-syncing as well as signs such as suspicious eye movements that indicate getting answers from other people.
“Beyond fraud mitigation, AI brings skills to the forefront and is able to remove subjectivity by eliminating bias during screening, basing decisions on fitness for a role, interest in a role, and skill match for the role, rather than formal credentials and subconscious bias that comes through the way people look or their network,” said Kumar.
What Future Applications of AI Can We Expect?
Science fiction movies postulate AI systems that become so smart they decide to exterminate humanity. Or AI that can comfortably take care of every single decision that needs to be made during a lengthy voyage to distant worlds.
Such capabilities are obviously far off. But, what can we expect over the near future?
Practical and deep reinforcement learning (RL)
Cronje named practical reinforcement learning, which leans into AI’s capacity to explore and exploit an industrial space. An example is work being done at DeepMind and Google in collaboration with a building management system provider. It demonstrates RL’s utility in helping to solve a real-world industrial problem (in this case, inefficient commercial cooling systems in data centers). By adapting RL to learn from offline data and within certain constraints, significant energy savings were achieved, Cronje added.
“Deep Reinforcement Learning is also being proposed as an adaptive solution at the leading edge of chip design to satisfy the growing demand for next-generation computing power,” said Cronje. “RL may be pivotal in assigning increasingly optimal locations for circuit components within the chip’s core area, meaning rapid development of future shaping semiconductors working in symbiosis with the hardware they power.
“Finally, on the bolder end of the AI spectrum, innovators from diverse fields are collaborating with AI specialists, so RL and simulation technology can leverage nuclear fusion as an inexhaustible source of clean energy.”
Natural language prompts for AI
Nagarajan suggested another near-future advance: Instead of using stock photos, users will describe the images they want and craft it to their liking with a few prompts. Rather than writing complex SQL queries, users will describe what they want in natural language to get the results that way. He thinks AI is going to shift a large amount of creative work into more hands.
“A new class of workers will not learn Excel formulas or PowerPoint layout techniques but will learn to command an AI with the right statements,” said Nagarajan. “The notion of programming will merely be communicating with a computer. This will unlock a step change in productivity growth that we haven’t seen since computers and email began to be commonplace.”
Understanding contextual risks for insurance and other uses
Banin added that much work is being done to raise the understanding and coverage of more risks in real time as well as the fusion of factors that go beyond driver behavior and surrounding context. These factors may include weather conditions, road conditions, surrounding vehicle and pedestrian traffic congestion, and relative riskiness of routes.
The next step is to apply this understanding not only for driver safety but also for underwriting insurance and improving fleet performance and efficiency.
“We continue to push the boundary on applying deep learning, convolutional neural networks to fuse understanding of more and more contextual risks, trigger actions that enable collision avoidance, and catalyze driver behavior change that reduces risk,” said Banin.
Nicolas Sekkaki, general manager of applications, data and AI at Kyndryl, foresees a near-horizon filled with responsible AI solutions that address trust, risk, ethics, security, and transparency. Additionally, he expects to see big improvements in solutions that target personalized insights, whether related to aspects such as credit risk or recommendation engines for dynamic pricing or influencing buying decisions.
Insurance companies, for example, will use AI to transform how cars, homes, businesses, and individuals are insured within more highly flexible programs. This focus will include dynamically and automatically adjusting coverage and making sure it’s optimized and personalized at any given moment.
AI solutions will monitor human behavior, proactively anticipate challenges and opportunities along the customer life journey and adjust coverage to provide contextual offers.
A large U.S. insurance carrier, for example, expects that within a few years, about 70% of the new car policies will be usage or behavior based. This will require insurance companies to leverage analytical models and machine learning algorithms as a basis for analyzing the vast amounts of information gathered from phone sensors and connected cars.
Augmented AI for human resources development
HR, too, will be a prominent beneficiary of future AI developmental efforts.
“Artificial intelligence will help businesses augment their entire HR life cycle from sourcing, assessing skills, interviewing, hiring, and mapping talent needs to keep existing talent happy and engaged,” said Kumar.
“In HR, recruiting and staffing, tasks like screening via chatbots, scheduling next steps, and ranking based on skill, fit, and interest are already happening. And remaining manual and repetitive tasks across every industry will eventually all go to a form of AI.”