Managing Complexity

Athough AI has great potential to help humans strategize beyond the limits of 86 billions neurons, it has many challenges including trust. Introducing the Foundation Model Transparency Index. Organizations investing in layering applications on these foundational models certainly need this.

From Sayash Kapoor on his site aisnakeoil.com. Sayash was recently included in the TIME 100 Most Influential People in AI.

Foundation models such as GPT-4 and Stable Diffusion 2 are the engines of generative AI. While the societal impact of foundation models is growing, transparency is on the decline, mirroring the opacity that has plagued past digital technologies like social media. How are these models trained and deployed? Once released, how do users actually use them? Who are the workers that build the datasets that these systems rely on, and how much are they paid? Transparency about these questions is important to keep companies accountable and understand the societal impact of foundation models.

2023 Nobel Prize awards go to those that shrink human knowledge.

Anne L’Huillier, Pierre Agostini and Ferenc Krausz shared the 2023 Nobel Prize in Physics for producing laser pulses lasting mere attoseconds.

One attosecond is one-quintillionth of a second, or 0.000000000000000001 seconds. More attoseconds pass in the span of one second than there are seconds that have passed since the birth of the universe.

Katalin Karikó and Drew Weissman share the Physiology or Medicine Nobel Prize for development of mRNA  to provide instructions to cells to make proteins (10 nanometers). A typical atom is 0.1 to 0.5 nm in diameter. DNA molecules are about 2.5 nanometers wide. A typical virus is about 100 nm wide. The work led to the development of Covid-19 vaccines (75 – 89 nM) administered to billions around the world.

Moungi G. Bawendi, Louis E. Brus and Alexei I. Ekimov are awarded the Nobel Prize in Chemistry 2023 for the discovery and development of quantum dots (1.5 – 10 nm). These tiny nanonanoparticles are essential for a wide range of applications including LED displays, solar cells, and biomedical imaging.

Expanding human knowledge of our attosecond and nanoscale world enables more innovation as well as more complexity to manage.

Interesting new Alzheimer’s research that looks beyond amyloid plaques buildup in the brain.

Researchers conducted the most extensive analysis on the genomic, epigenomic, and transcriptomic changes in Alzheimer’s patient brains. By analyzing over 2 million cells from 400 postmortem samples, they offer insight into the interplay of four areas to help treat Alzheimer’s disease. 

Transcriptome – RNA-sequencing to analyze the gene expression patterns of 54 types of brain cells. They found impairments in the expression of genes involved in mitochondrial function, synaptic signaling, and protein complexes needed to maintain the structural integrity of the genome.

Epigenomics – The chemical modifications that effect gene usage within a given cell. The found they occur most often in microglia, the immune cells responsible for clearing debris from the brain.

Microglia – brain cells that make up 5 to 10 percent of the cells in the brain. They clear debris from the brain, are immune cells that respond to injury or infection and help neurons communicate with each other. They found as Alzheimer’s disease progresses, more microglia enter inflammatory states, the blood-brain barrier begins to degrade, and neurons begin to have difficulty communicating with each other.

DNA damage – during memory formation, neurons create DNA breaks. These breaks are promptly repaired, but the repair process can become faulty as neurons age. They found that as more DNA damage accumulates in neurons, it gets more difficult to repair the damage, leading to genome rearrangements and 3D folding defects.

The hype and doom of generative AI models such as ChatGPT often leave out the work ahead before their predictions can happen. While these new large language models will be transformative, they will have to address numerous technical, social and economic challenges.

Arjun Ramani and Zhengdong Wang describe many of these challenges in Why transformative artificial intelligence is really, really hard to achieve on the site The Gradient.

They discuss the challenges of AI creating productivity gains in areas such as healthcare, education, and construction that is labor intensive. The fine motor control of robots are not advancing as fast as large language models that still require millions of humans to annotate and train these AI models. They remind us that a lot of process knowlege (that AI models will use) is not written down anywhere citing Michael Polanyi saying “that we can know more than we can tell.”.

There are many AI challenges including data quality with common ontologies, securing data quanity, data sources (local vs. industry), model reproducibility, caputuring inputs and outputs that might not be digitized, privacy, data rights, trust, transparency and transformation of current processes. I have no doubt these will get addressed. How they get addressed will determine the promise and perils of AI.

Navigating the rules of the health insurance can be complex.

A majority of insured adults (58%) say they have experienced a problem using their health insurance in the past 12 months – such as denied claims, provider network problems, and pre-authorization problems.

This is from a recent Kaiser Family Foundation (KFF) survey of 3,605 U.S. adults with health insurance. It also found:

Nearly half of insured adults who had insurance problems were unable to satisfactorily resolve them, with some reporting serious consequences.

Yet the study found most viewed their health insurance favorably. Those with good health rated it higher than those reporting poor health.

Ben Dickson sheds light on what is behind ChatGPT and Microsoft’s Bing Chatbot named Sydney. These large language models (LLMs) will likely find their way into our normal routines. It is important for us to understand their strengths and weaknesses before is gets your coffee order wrong and eloquently explains to you why it was right.

His well-written post “To understand language models, we must separate “language” from “thought” describes what these LLMs, created through machine learning of massive data sets, do well and what they struggle with. He cites a recent paper Dissociating language and thought in large language models: a cognitive perspective that found:

LLMs show impressive (although imperfect) performance on tasks requiring formal linguistic competence, but fail on many tests requiring functional competence.

Leadership can be a lonely place. Especially during the Covid-19 pandemic if you are restaurant owner, school superintendent, healthcare leader, or wedding planner. It is impossible to process the complexity of the many knowns, unknowns and uncertainties to forecast the future and satisfy the impacted people. Yet, that is what we ask them to do.

As an example, here are 15 Covid-19 forecasts healthcare leaders must get right. See more.

The global pandemic has forced a curriculum change to understanding the complexity of our modern life. We were auto enrolled in immunology, epidemiology and forecasting lessons that are prerequisites to make sense of therapeutics, vaccines, and virus transmission based on human behavior. This understanding will help us forecast R value infection rates, vaccine timelines, eventual herd immunity and a new date for the family reunion.

The brilliant and talented Atlantic writer Ed Yong provides a ‘must read’ lesson on immunology and COVID-19, The Pandemic’s Biggest Mystery Is Our Own Immune System. Ed is a great explainer. If you like this article, I would strongly recommend you read I Contain Multitudes: The Microbes Within Us and a Grander View of Life. He explains the complexity of science in a fun and easy to read way. It is one of my favorite books.

The past six months has provided us an education in epidemiology, virology, and human behavior. While we didn’t sign up for these lessons, they became prerequisites for making sense of living during a pandemic and predicting what’s next. How will our kids be educated? Will elections be safe and fair? Will the newly unemployed be hired? Will the entertainment businesses survive?

We are seeing signs of coronavirus outbreaks slowing in Arizona, Texas and Florida. This is likely the result of human behavior changes in staying home, wearing masks, hand washing, and social distancing. We are also learning valuable insight to how our immune systems combats the coronavirus. This new insight may not impact the fall, though it may have a major impact on 2021.

Our National Institute of Health (NIH) Director Dr. Francis Collins highlighted this week new insight on how our T cells fight COVID-19. While the antibodies produced by our immune systems get most of the media attention related to vaccines, therapeutics and testing, the human T lymphocyte (T cell) needs to be added to our vocabulary and discussions.

The findings of a new study reported in Nature, suggests these T cells may protect some people newly infected with SARS-CoV-2 by remembering past encounters with other human coronaviruses. Four of the six known coronaviruses are responsible for the common cold. This might explain why some people get very sick while others are asymptomatic.

This study indicates we need to add T cells to the general publics epidemiology and virology curriculum. T cells may soon become part of our testing, therapeutics and vaccines discussions.

When it comes to managing the complexity of living during a pandemic,
developing predictions based on our understanding of many impacting elements is essential. These predictions will become the foundation for our strategies to new manage the complexity of 2021.

Image: Scanning electron micrograph of a human T lymphocyte (T cell) from a healthy donor’s immune system from National Institute of Allergy and Infectious Diseases/NIH.