Machine Learning: The Era of Artificial Intelligence in Our Industry – 101
We’ve heard about Artificial Intelligence (AI) for decades and now it’s become more prevalent in the regenerative medicine space. Developing effective regenerative therapies requires the ability to analyze large amounts of complex data, which is where AI enters the picture. Machine learning empowers researchers to analyze vast amounts of data, recognize patterns, make predictions based on that data, and learn from their mistakes and adjust their behavior accordingly without having to be programmed in [1] . If you’ve been wondering where to start, we’re bringing the basics in a series of our blogs on AI and how it relates to our industries (e.g., biologics, cellular therapies, medical device, drug therapy, etc.). Let’s dive in!
What are some of the fundamentals of AI?
Concepts for understanding machine learning, deep learning, and neural networks
Reasoning, problem-solving, and perception
Natural language processing
Ethical considerations
The capabilities of AI can be understood as a spectrum, categorized generally by the level of intelligence and adaptability systems demonstrate. These are the key levels [2] :
Artificial Narrow AI (a.k.a. Weak AI) is the only type of AI that exists today. This type of AI is trained to perform a narrow or specific task without possessing general intelligence. This type cannot perform outside its defined task and targets a single subset of cognitive abilities and advances in that spectrum (e.g., Siri, Amazon’s Alexa, IBM Watson and OpenAI’s ChatGPT).
General AI (a.k.a. Strong AI) refers to a theoretical concept that possesses the ability to understand, learn, and apply knowledge across a wide range of tasks without the need for humans to train the underlying models. This allows General AI to learn and perform at a human-like level of intelligence.
Superintelligent AI (a.k.a. Artificial Superintelligence) is also a theoretical concept and, if ever realized, goes beyond human intelligence and capabilities (it would have evolved beyond the point of understanding human sentiments and experiences to feel emotions, have needs, and possess beliefs of its own. This raises ethical and philosophical questions about the future development of AI).
Machine Learning is the foundation involving algorithms that allow systems to learn from data, identify patterns, and make informed decisions. Macine Leaning is a subset of AI that enables systems to learn from data (input) without being explicitly programmed for every task. Algorithms analyze large datasets to recognize patterns and extract valuable insights, allowing machines to improve their performance over time. These algorithms can be trained by using techniques such as [3] :
Supervised learning is training the algorithm on labeled data where it learns to make predictions or classify inputs based on outputs that are already provided, like identifying spam emails.
Unsupervised learning is training the algorithm on unlabeled data to discover hidden patterns and relationships independently. This type is ideal for clustering tasks.
Reinforcement learning focuses on training to rely on trial and error, where the model is rewarded when it makes desirable actions (depending on the nature of the task and the available data). This is applied in fields like robotics, gaming, and autonomous systems.
Neural networks based inspirationally by the human brain, consist of interconnected layers of nodes or artificial neurons (input layer, one or more hidden layers, and an output layer) that process information and learn to recognize patterns through training and improving their accuracy over time [4] . Additional considerations for neural networks:
Computational models inspired by the structure and function of the human brain’s interconnected neurons.
Comprised of layers of artificial neurons or nodes that process input data through the interconnected layers to generate output.
Each neuron receives input signals, applies a transformation using weighted connections, and passes this result to the next layer.
Through ‘training’, neural networks adjust the weights of these connections to learn patterns and relationships in the data.
This process allows the networks to perform tasks such as classification, regression, and pattern recognition with remarkable accuracy.
Deep Learning is a subset of machine learning that uses complex artificial neural networks with many layers to analyze vast amounts of data and perform highly sophisticated tasks [1] . These concepts allow AI to process data, learn patterns, and make intelligent decisions.
Deep layers are capable of automatically learning intricate patterns and representation in large and complex datasets making them powerful for tasks such as image recognition, natural language processing and speech recognition.
Unlike traditional models with a few layers of neurons, deep learning can consist of dozens or even hundreds of layers, enabling them to learn complex hierarchical representations of data and improve over time.
Machine learning is used in some applications such as:
In tissue engineering, AI can be used in analyzing physiochemical and biological properties of a wide range of materials. Algorithms can identify patterns and associations in cellular behavior and interactions, thereby predicting cell behavior in various environments which is important in designing and optimizing engineering of functional organs and tissues [1] .
AI driven healthcare tools to support physicians by interpreting medical images and diagnosing disease, thereby providing a more accurate and timelier treatment [3] .
Performing computational simulations in medical applications to lower costs and provide faster results as compared to clinical and laboratory methods [1] .
Analyzing large datasets and identifying patterns that could be missed by humans as well as helping researchers better understand the underlying disease mechanisms and develop more effective therapies to address them [1] .
For drug discovery, AI can analyze large datasets to detect patterns and associations by analyzing chemical structures and properties to help identify potential drug candidates, assist in validating drug targets, predict the toxicity of potential drug candidates and therefore identify potential safety concerns early in the discovery process.
While there are benefits to AI in our industries, with it comes limitations. The quality, quantity, and availability of data may be limited where patient data is minimal or diverse, thus making it challenging to train AI models effectively. Another limitation is the complexity of biological systems such as cell therapies that involve intricate interactions between cells and tissues, making the analysis difficult for many of the machine and deep learning algorithms to model them accurately [1]. As AI continues evolving and being integrated into our industries, we look forward to better ways of overcoming challenges and increasing efficiencies!
Stay tuned as we continue learning more about the basics of artificial intelligence!
References
[2] https://www.ibm.com/think/topics/artificial-intelligence-types