This blog is a continuation of the Building AI Leadership Brain Trust Blog Series which targets board directors and CEO’s to accelerate their duty of care to develop stronger skills and competencies in AI in order to ensure their AI programs achieve sustaining results.
My last two blogs introduced value of science and stressed its importance to AI, and focused on the importance of AI professionals having some foundation in computing science as a cornerstone for designing and developing AI models and production processes, as well as the richness of complexity sciences and appreciating that integrating diverse disciplines into complex AI programs is key for successful returns on investments (ROI).
This blog introduces the importance of physics and explores its relationship to AI as often I see AI solutioning teams missing physics skills in the solutioning constructs – which I believe is a strategic mistake for many complex AI programs. Its important for C levels to understand that AI is not a singular discipline it requires many other skills to get the solution architecture right. So deeply understand the business problem in front of you – and the more complex the problem is the increased value physicists will have in guiding you forward.
Atomic molecule on blackboard
In the Brain Trust Series, I have identified over 50 skills required to help evolve talent in organizations committed to advancing AI literacy. The last few blogs have been discussing the technical skills relevancy. To see the full AI Brain Trust Framework introduced in the first blog, reference here.
We are currently focused on the technical skills in the AI Brain Trust Framework
MORE FOR YOU
1. Research Methods Literacy
2. Agile Methods Literacy
3. User Centered Design Literacy
4. Data Analytics Literacy
5. Digital Literacy (Cloud, SaaS, Computers, etc.)
6. Mathematics Literacy
7. Statistics Literacy
8. Sciences (Computing Science, Complexity Science, Physics) Literacy
9. Artificial Intelligence (AI) and Machine Learning (ML) Literacy
What is the relevance of physics to AI as a discipline?
There are so many aspects of physics that can be applied to AI hence, it does not take one long to appreciate the value of this science discipline. One of the most significant discoveries in physics was the Higgs Boson Particle, often referred to as the God Particle which was discovered using an AI neural network to help identify complex patterns in particle collisions.
The last blog stressed the importance of complexity science and the most important aspect of physics is that this discipline teaches you about how to understand and decompose complex processes.
In prior blogs, I stressed that the importance of building an AI model requires three main enablements: 1) collecting and analyzing the data 2.) developing the AI model and 3.) evaluating the model outcomes and determining value. Each of these areas has relevance to physics and a strong AI expert will appreciate the value that physics know-how can bring to enable engineering teams to tackle the most complex problems in the world.
Let’s start first with data analysis. There are many forms of machine learning approaches, but the one that has the closest linkages to physics is neural networks which are trained to identify complex patterns, as well as find new patterns. Examples of how AI can be applied to solve a physics problem would be to classify thousands of images and be able to identify black holes, being able to detect subtle changes in light around objects is an example of the disciplines coming together.
Physics professionals also use terms like gravitational lensing for image analysis using neural networks to tease out the classifications to finer levels of details, while AI specialists simply say image processing. What is always a challenge in diverse disciplines is geek speak often confusing business leaders who cannot decipher the language meanings.
In addition, many acclaimed physicists purport that they are the major contributors to advancing the AI field, so rivalry friction exists in these disciplines as well, and pardon the pun.
Neural networks are particularly good at enabling AI models to be able to detect changes in radio waves or even earth’s gravitational waves, or to determine when specific rays may the hit earth’s atmosphere and provide timing insights as well.
Being able to encode different particle behavior and observe their subtle changes over time provides a rich bed of AI modelling analysis and interpretability for physicists to have deeper mathematical calculation insights to encode their observations more accurately.
Other physics terms that underlie neural networks include: compressibility or conductivity. What is even more exciting in bringing these two disciplines together is the area of quantum tomography, which equates to measuring the changes in a quantum state which has innovation relevance to quantum computing. Tomography is an exciting field which analyzes images by sections or sectioning through the use of any kind of penetrating wave. This method is used in diverse areas including: radiology, atmospheric sciences, geophysics, oceanography, plasma physics, astrophysics, quantum information, and other science areas. Its applications are endless and very exciting.
Machine learning methods help to advance physics, as well as physics has value and relevance to machine learning. The high computational value of machine learning is allowing physicists to tackle even more complex problems, like in simulating global climatic change leveraging geometric surfaces and applying deep learning onto curved surfaces.
An Imperial College computer scientist, Michael Bronstein and his researchers, helped to advance geo-metric deep learning methods and determined that going beyond the Euclidean plane would require them to reimagine one of the basic computational procedures that made neural networks so effective at 2D image recognition in the first place. This procedure lets a layer of the neural network perform a mathematical operation on small patches of the input data and then pass the results to the next layer in the network.
Without going into too many details these researchers re-imagined these approaches and recognized that a 3D shape bent into two different poses – like a bear standing up or a bear sitting down – were all instances of the same objects vs two distinct objects.
Hence the term Convolutional Neural Networks (CNN) was born. This type of network specializes in processing data in a grid like topology, such as an image, and each neuron works in its own receptive reference field and is connected to other neurons in a way that they cover the entire visual field, so after analyzing thousands of images of a cat or a dog this problem is not as difficult as there is easy access to this data set.
CNNs can detect rotated or reflected features in flat images without having to train on specific examples of the features and spherical CNNs can create feature maps from data on the surface of a sphere without distorting them as flat projections. The applications are endless and very exciting to physicists where object surface detection is key in their research methods.
Unlike finding cancerous tumors from diverse lung photos, finding medically accurate, quality labelling validated is a more difficult challenge to achieve.
In a government and academic research project they used a convolutional network (CNN) to detect cyclones in the data using newer gauge CNN detection method which was able to detect cyclones at close to 98% accuracy. A gauge CNN would theoretically work on any curved surface of any dimensionality The implications for climate monitoring using physics and AI techniques is unprecedented with these advancements.
In summary, both physics and machine learning have some similarity. Both disciplines are focused on making accurate observations and both build models to predict future observations. One of the terms that often physicists use is co-variance which means that physics should be independent of which kind or rule is used or what kind of observers are involved which nets out to simply stresses independent thinking.
Einstein stated this best in 1916 when he said: “The general laws of nature are to be expressed by equations which hold good for all systems of coordinates.”
Analyzing diverse patterns
What key questions can Board Directors and CEOs ask to evaluate their depth of physics linkages to artificial intelligence relevance?
1.) How many resources do you have that have an undergraduate degree in physics versus a masters degree or a doctoral degree?
2.) Of these total resources trained in physics disciplines, how many also have a specialization in Artificial Intelligence?
3.) How many of your most significant AI projects have expertise in physics to ensure increased inter-disciplinary knowledge know-how?
4.) How many of the Board Directors or C-Suite have expertise in physics with a knowledge blend of AI to tackle the worlds most complex business problems?
These are some starting questions above to help guide leaders to understand their talent mix in appreciating the value of diverse science disciplines to augment the AI solution delivery teams in enterprises.
I believe that board directors and CEOs need to understand their talent depth in science disciplines in addition to AI disciplines to ensure that their complex AI programs are optimized more for success. The last three blogs, including this one looked at three disciplines 1) Computing Science 2.) Complexity Science and this one on Physics – all written to reinforce the important that science disciplines are key to ensuring AI investments are successful, and continued investments are made to help them evolve and achieve the value to support humans in augmenting their decision making, or improving their operating processes.
The next blog in this AI Brain Trust series will discuss a general foundation of the key AI terms and capabilities to provide more knowledge to advance the C-Suite to get AI right and achieve more sustaining success.
To see the full AI Brain Trust Framework introduced in the first blog, reference here.
To learn more about Artificial Intelligence, and the challenges, both positive and negative, refer to my new book, The AI Dilemma, to guide leaders foreward.
If you have any ideas, please do advise as I welcome your thoughts and perspectives.