b0hr.ai: understanding language at depth
"What is it that we humans depend on? We depend on our words... Our task is to communicate experience and ideas to others. We must strive continually to extend the scope of our description, but in such a way that our messages do not thereby lose their objective or unambiguous character …
Quoted in Philosophy of Science Vol. 37 (1934), p. 157, and in The Truth of Science: Physical Theories and Reality (1997) by Roger Gerhard Newton, p. 176
Today, October 7th, would have been Niels Bohr's 134th birthday. Niels Bohr was a Danish physicist and philosopher who bridged the evolution of physics from classical and absolute to quantum and relativistic. He did this by embracing multiple schools of thought at once with the idea of *complementarity*, best known from his proposal that we can think of light as *both* wave and particle.
Inspired by his life's work and in honor of his memory, doc.ai is launching a new division of the company, b0hr.ai, an initiative to invent new approaches to natural language systems in the field of healthcare and medical research.
We're starting b0hr.ai because we're aware that current natural language technologies, dependent on huge datasets and opaque models, aren't sufficient for the emotionally and ethically demanding tasks of healthcare. Simply put, the bar for developing AI in healthcare is higher.
We believe that human language processing, whether reading, listening, speaking, or writing, always involves improvisational learning. While there are rules, exemplars, paradigms and brain structures that shape language, our interpretation and generation always involves the recognition or invention of variations and the carrying of those variations forward into the future to varying degrees.
Take the name of our site, “b0hr.ai”. Readers familiar with Bohr can read the meaning “easily”, influenced by both what they know by the shapes of letters in Western alphabets. In a more common example, imagine a spoken conversation involving different accents. Despite the differences, conversants usually can readily adapt and learn to understand one another’s words despite differences in pronunciation or rhythm.
This is not remarkable or extreme but commonplace; it is how human beings understand whether the variations are intended or accidental. It is not, however, the way that most computers approach language, whether they’re using programmed rules and grammars or large training corpora.
The past decades’ breakthroughs in AI have arisen from the combination of big data and big compute allowing simple approximate numeric models to be trained on immense subsets of the infinite space of variations. And many of the flaws with these models, from crazy translations to unforeseen bias, arise from these approximations.
b0hr.ai will develop new theories, tools, and practices for natural language based on these kinds of inference and learning and making systems that are more transparent, correctable, and personal. This will be crucial as healthcare continues to grow more socially, technically, and scientifically personal. We will be continually integrating research with tool development with real-world prototypes and deployments.
Our RAD (Research & Advanced Development) roadmap will constantly evolve but will include:
Hybrid engineered/trained architectures to combine explicit and implicit insights about tasks and domains;
Multi-model inference & learning which assembles a plurality and variety of models for explainability and quick adaptability;
Testing and analysis of trained models to better understand their scope and limits;
Conversational fluency, using individual and collective history, to both streamline and repair conversational engagements.
b0hr.ai is actively hiring and also seeking compelling use cases to explore with doc.ai’s world-class solutions and deployment teams. Connect with us here.