Investment and interest in AI is expected to increase in the long run since major AI use cases (e.g. autonomous driving, AI-powered medical diagnosis) that will unlock significant economic value are within reach. These use cases are likely to materialize since improvements are expected in the 3 building blocks of AI: availability of more data, better algorithms and computing.
Short term changes are hard to predict and we could experience another AI winter however, it would likely be short-lived. Feel free to jump to different sections to see the latest answers to your questions about the future of AI:
Will interest in AI continue to increase?
Short answer: Yes.
Interest in AI has been increasing
According to AI Index, the number of active AI startups in the U.S. increased 113% from 2015 to 2018. Thanks to recent advances in deep-learning, AI is already powering search engines, online translators, virtual assistants and numerous marketing and sales decisions.
The Google Trends graph below shows the number of queries including the term “artificial intelligence”. Since 2015, popularity of AI grew to 2-3x the level in 2015.
One notable exception is 2019. During 2019 interest in AI seems to have stabilized. There could be a stabilization of interest in AI in the short term.
There are high value AI use cases that require further research
Autonomous driving is one popular use case with an increasing trend. As Tesla and Audi manufacture semi-autonomous vehicles today, they still require drivers to control. This technology rapidly continues to improve to reach a fully automated driving level. McKinsey predicts that roughly 15% of vehicles sold in 2030 will be fully autonomous.
Another use case is the conversational agents. We commonly encounter with AI agents in customer services and call centers. However, capabilities of these agents are currently quite limited. As AI research progresses, conversational agents will improve to handle almost all customers’ tasks in the future.
AI research effort continues to grow
Between 1996 and 2016, the number of published papers on AI has increased eight times, outpacing the growth in computer science papers.
Research may need to continue in new directions beyond deep learning for breakthrough AI research. There are AI researchers like Gary Marcus who believe that deep learning has reached its potential and that other AI approaches are required for new breakthrough. Gray outlined his observations on the limitations of AI in this paper, answered most critical arguments against his paper and put a timeline on this predictions. He expects VC enthusiasm in AI to be tempered in 2021 but expects the next AI paradigm unlocking commercial opportunities (e.g. the new deep learning) to be available some time between 2023-2027.
What are the key trends that shape the future of AI?
AI systems so far relied on these for improvement: availability of more data, better algorithms and computing. In all 3 areas, there is potential for dramatic improvements though it is hard to put these against a timeline.
Advances in computing power
Deep learning relies on computing power to solve more complex problems. With current technology, learning may take too long time to be beneficial. Therefore, there is need for advances in computing power. With new computing technologies, companies can have AI models that can learn to solve more complex problems.
Even the most advanced CPU may not improve the efficiency of an AI model by itself. To use AI in cases like computer vision, natural language processing, or speech recognition, companies need high-performance CPUs. AI-enabled chips become a solution to this challenge. These chips make CPUs “intelligent” for optimizing their tasks. As a result, CPUs can work for their duties individually and improve their efficiency. New AI technologies will require these chips to solve complicated tasks and perform them faster.
Companies like Facebook, Amazon, and Google are increasing their investments in AI-enabled chips. Below you can find a chart of global equity funding for AI-enabled chip startups.
These chips will assist next-generation databases for faster query processing and predictive analytics. Industries like healthcare and automobile heavily rely on these chips for delivering intelligence.
Advances in GPUs
GPUs are one of the most commercially used type of AI enabled chips.
Rendering an image requires simple computing power but needs to be done on a large scale very quickly. GPUs are the best option for such cases because they can process thousands of simple tasks simultaneously. As new technologies in GPU renders better-quality images since they do these simple tasks a lot faster.
Modern GPUs have become powerful enough to be used for tasks beyond image rendering, such as cryptocurrency mining or machine learning. While CPUs usually used to do these tasks, data scientists discovered that these are repetitive parallel tasks. Thus, GPUs are widely used in AI models for efficient learning.
Traditional computer systems work with binary states; 0 and 1. However, quantum computing takes this to another level and works with quantum mechanics. This enables quantum systems to work with qubits, instead of bits. While bits consist of 0 and 1, qubits consist of 0, 1 and an additional state, which includes both at the same time. This additional state enables quantum computing to be open to new possibilities and provide faster computation for certain tasks. These tasks include neural network optimizations and digital approximations.
IBM states that it will be possible to build a quantum computer with 50-100 qubits in the next 10 years. When we consider that the 50-qubit quantum computer works faster than today’s best 500 supercomputers, there is significant potential for quantum computing to provide additional computing power.
Advances in data availability
This is a point that does not need to be explained in much detail. Data availability has been growing exponentially and is expected to continue to do so with increasing ubiquity of IoT devices.
Advances in algorithm design
While the capabilities of AI improve rapidly, the algorithms behind AI models will also evolve. The advancements in the algorithm designs will enable AI to work more efficiently and be available to more people with less amount of technical knowledge. Below you can find the prominent advancements in AI algorithm designs.
Explainable AI (XAI)
One of the main weak points of AI models is its complexity. Building and understanding an AI model requires a certain level of programming skills and, it costs time to digest the workflow of the model. As a result, companies usually benefit from the results of AI models without understanding their workflow.
To solve this challenge, Explainable AI makes these models understandable by anyone. XAI has three main goals:
- How the AI model affects developers and users
- How it affects data sources and results
- How inputs lead output
As an example, AI models will be able to diagnose diseases in the future. However, doctors also need to know how AI comes up with the diagnosis. With XAI, they can understand how AI makes its analysis and explain the situation to their patients accordingly.
Transfer learning is a machine learning method that enables users to benefit from a previously used AI model for a different task. In several cases, it is clever to use this technique for the following reasons:
- Some AI models aren’t easy to train and can take weeks to work properly. When another task comes up, developers can choose to adopt this trained model, instead of creating a new one. This will save time for model training.
- There might not be enough data in some cases. Instead of working with a small amount of data, companies can use previously trained models for more accurate results.
As an example, an AI model that is well-trained to recognize different cars can also be used for trucks. Instead of starting from scratch, the insight gained from cars will be beneficial for trucks.
Reinforcement learning (RL)
Reinforcement learning is a subset of machine learning which aims AI agent to take action for maximizing its reward. Rather than traditional learning, RL doesn’t look for patterns to make predictions. It makes sequential decisions to maximize its reward and it learns by experience.
Today, the most common example of RL is Google’s DeepMind AlphaGo which has defeated the world’s number one Go player Ke Jie in two consecutive games. In the future, RL will also be available in fully automated factories and self-driving cars.
Advances in AI building tools
Though these are not novel algorithms, they can reduce the time to build models and enable both AI research and commercialization
Neural network compatibility and integration
Choosing the best neural network framework is a challenge for data scientists. As there are many AI tools in the market, it is important to choose the best AI tool for implementing the neural network framework. However, once a model is trained in one AI tool, it is hard to integrate the model into other frameworks.
To solve this problem, tech giants like Facebook, Microsoft, and Amazon are cooperating to build Open Neural Network Exchange (ONNX) to integrate trained neural network models across multiple frameworks. In the future, ONNX is expected to become an essential technology for the industry.
Automated machine learning
AutoML supports companies to solve complicated business cases. With this technology, analysts won’t need to go through manual machine learning training processes. They can even evolve new models that can handle future AI challenges. As a result, they will focus on the main case instead of wasting time for understanding the workflow.
AutoML also offers customization for different business cases. This enables flexible models when you combine data with portability. To learn more about AutoML, you can check our article.
What are the future technologies to be enabled by AI?
AI use cases will shape the development of AI. Availability of capital depends on use cases and more valuable use cases would motivate companies and government to invest more.
The improvement of AI will make our intelligent systems even more intelligent. Our cars will drive themselves, houses will adjust their electricity usage, and robots will be able to diagnose our illnesses. In other words, AI will cover more in our lives and will automate our daily tasks. Here are a few use cases of AI technologies that currently either exist in quite limited functionality or limited scope (research projects). Improvement of these technologies will unlock significant value.
- AI assistants
- AI-based medical diagnosis
- Autonomous payments
- Autonomous vehicles
- Bionic organs
- Conversational agents
- Smart cities
- Smart dust
Cloud computing based use cases
Cloud computing aims to create a system where you can achieve computing functions whenever you want. According to Gary Eastwood from IDG Contributor Network, cloud computing and AI will fuse in the future.
The integration of AI will help AI models to access information from the cloud, train themselves and applies new insights into the cloud. This enables other AI models to learn from these new insights. This fusion improves calculation power and the capability of treating many data and intelligence.
The possible use cases of cloud computing include AI-lead drones, sensor networks, and smart dust.
Extended Reality (XR)
Besides technologies like Virtual Reality or Augmented Reality, start-ups are experimenting with bringing touch, taste, and smell to enhance these immersive experiences with the support of AI technologies. While XR may bring several security issues in the future, XR will be essential to improve worker productivity and the customer experience in the future.
According to Accenture, the designers at Volkswagen can experience the car’s look, feel and drive—spatially, in 3D—thanks to XR tools.
Convergence of IoT and AI
Another trending technology IoT will merge with AI technologies in the future. AI can be used in IoT platforms in use cases like root cause analysis, predictive maintenance of machinery or outlier detection. Devices like cameras, microphones, and other sensors collect this data from video frames, speech synthesis, or any other media. Then, it is trained in the public cloud environment with advanced AI technologies based on neural networks.
How useful was this post?
Click on a star to rate it!
Average rating / 5. Vote count:
What did you like about this post?
Appreciate it if you leave your name and surname so we can publish your review as a testimonial
Thank you for your feedback!