AI development expertise for life sciences & Biotech

---

  AI has of late been widely-applied beyond the initial limited areas of a) image enhancement, 

b) text searching, matching, analysis and synthesizing semi-intelligible output, and c) robotics (learning successful movement)

One of the main weaknesses of AI otherwise, and reliance on its' conclusions or output--- is the quality, correctness, completeness and accuracy/reliability of its' input.

This crucial step does not change, whether human or AI does the compilation, distilling, analysis, organizing and coding (setting data elements):

Unfortunately, humans have partially skipped this crucial step, not really taking the time and extreme effort to determine and codify the indicia, characteristics and actual reliability of source information and documents, particularly here in the subject areas of advanced science.

Be very selective as to exactly what an AI application is going to use for input; For the forseeable future, raw data, papers and the internet have very limited usefulness and questionable quality. 

For example, ChatGPT was not connected to the internetsince 2021 to 2024 and now only with mobile in a limited manner. Yet users were excited and using it as if it was.

Too many "experts" designing, creating, writing, training and testing AI may minimize or partly disregard, or even prefer to not draw attention to this near-intractable problem. So many in this newly-booming field rapidly see visions of dollar signs, as markets are "throwing money" at AI and its' "infrastructure" and the companies beginning to exploit the newer capabilities.

That is, in every new technology (recall the internet bubble of the late 1990's), the hype, hope and expectations can get far ahead of reality.

Our cutting-edge drug discovery platform uses advanced molecular modeling and artificial intelligence to identify novel drug targets and accelerate the development of life-saving therapies.

The reality that endangers the new AI boom--- remains the huge problem underlying much of scientific and other progress: the lack of very highest-level testing, "rating", screening and classifying of existing and on-going research (input) in each field. The peer review system does not fully test the reality of many of the hypothesis and conclusions of much of the research. The AI itself can not be relied upon to independently evaluate, set up the labs and properly run the experiments to validate results, or form groups of top scientists to simply rate the papers in accord their best assessments.

Search engines began limited progress toward rating quality of papers, but the method is imperfect by the external research limitations, environment and other factors, as herein:

Popularity ratings from citations may not be the best manner of selection.  We've seen lesser Journal publishers not have the integrity to halt publication of disavowed, discredited  or infringed works. They make money from each submittal that is published, and only yield by a legal challenge.  The infringement issue looms large in the AI future.

There's a huge problem without a proper decision tree heirarchy with sources/ citations.

Our state-of-the-art facilities and experienced team enable us to produce high-quality biologics and cell therapies at scale, helping to bring innovative treatments to market faster.

Vetting and rating research may be done Internally to some companies: These validation experiments and process/production testing may be being done, but narrowly for their industry, segment and products. That is, companies have no interest and can not legally refine their proprietary technologies and decisions and then tell everyone (their competitors) about what really works, what multiplies yields, and what fails.

The other obstacles both in industry and academia are both competition and incomplete disclosure of ambiguity of key steps or methods (deliberately or oversight). And finally, top scientists see flaws in papers, theories, methods and experiment design... But there is no incentive to talk about these or challenge these mistakes/oversights, unless it is a competitor's paper.

 You have a "craze" AI bubble bumping up against systemic obstacles (next)

So there are obstacles, plus  stock market AI "irratinoal exhuberance". ( to quote  Alan Greenspan in other prior bubbles).  (This is not advice.  Bubbles can run for years).

On the other hand, it's a tool, and investors and entrepreneurs know "look for the picks and shovels producers" (an old reference on the main ones who won out in the Gold Rush).

Another adage is.. "go in around the edges of something big".

Micron memory, energy storage, NVidia, AI.  See a pattern here?

Growth industries matter. 

Which leads us back  to other limitations on real progress by obstacles like regulation (must use mouse studies (example)) and politics (in and out the science field and in business).

There is tremendous corporate culture inertia often holding back innovation, sometimes based on politics, social initiatives and incompetence now becoming a big concern of shareholders.

Also there is "cultural inertia" and resistance to change, as in health lab tests, classifying cancer cells and stages, standards, insurance codes, etc. ( "It's always been done this way and we believe it works.")

This extends to large government agencies and professional associations limiting what can be "approved".

Then the legal claims blossom.. over copyrights, IP and plagerism.

The methodolgies of training, testing and especially using (not very smart) have created huge artificial limits.  The public tools like Google..Can we say out loud?  don't seem very good, and increasingly hedged, not so relevant.
 

Also markets experiencing "irrational exuberance" to invest-- add big pressures to skip the huge requirement to make sure the "input" has no "garbage" and is properly rated, classified, screened, reviewed and codified.
 

Not least is that persons who have worked with computers over their evolution of the last 40 years know that ultimately some standards must be established to designate, standardize and codify (name the fields, indexing, data structures) key information, similar to what everyone must do to operate highly-structured spreadsheets.

Ultimately users want conclusions, summaries, exceptions reports: As in business management, the most important information is distilled into summaries, exception reports (what departments ran large overtime? What yield or failures came from key departments?), totals, year on year comparatives, etc. 

  Legacy and "expert systems .. do all that extremely well.  Remember, the German's lost WWII

why?  The answer is instructive and relevant.