Posts contrassegnato dai tag ‘IBM Watson’

Cognition-as-a-Service

Pubblicato: 13 gennaio 2014 da Paolo Magrassi in Luoghi comuni
Tag:, ,

watsonWatson si sta allargando dalla Sanità (dove presto cambierà le carte in tavola, vedrete) al customer service.

Il Watson Engagement Advisor aiuta gli operatori di call center -e anche direttamente i clienti- mettendo a disposizione risposte tratte dalla sua immensa facoltà di online content analytics, unita alla capacità di intendere il linguaggio naturale.

In entrambi i campi, Watson è immensamente più bravo del 95% delle persone.

Sta dunque per tramontare l’era dei poveri diavoli malpagati e maltrattati i quali, all’oscuro di tutto e dotati di mezzi informatici preistorici (in Italia, quasi assenti), dovrebbero rispondere su argomenti che vanno dall’automotive alla finanza, dall’informatica alla Grande distribuzione organizzata.

I clienti smart potranno abbeverarsi direttamente ad “ask Watson” nella Nuvola. E ci dispiace per gli altri… 🙂

I posti di lavoro che W decimerà potranno essere in parte riciclati per fargli il data entry, in parte per trarne suoi istruttori / addestratori.

Paul Allen (Microsoft’s co-founder) and Mark Greaves wrote on Technology Review about a week ago that the singularity is not near because we are a long way away from understanding how human cognition works.

I had been thinking the same from 1975, when I first became interested in AI, to approximately 2001.

At about that time I realized that the quest for human-like intelligence (in, e.g., game playing, bioinformatics, robotics, creative work) was no longer being pursued by imitation but rather by brute force.

Nowadays, the tools of the trade for building “intelligent” software are much less neuron-like chips and logic programming (even though these technologies are still present), and much more

  • very large databases
  • untra-fast pattern recognition
  • cellular automata / agent-based systems

These are the grounds on which some of the most impressive intelligent technologies are firmly based , including, to name the popular ones, Deep Blue, Watson, or Google Translator.

The logic programming, inevitably mimicking human way-of-reasoning, is conceptually the same as 40 years ago, when neuroscience was not as developed as today and we knew less about the human brain. But hardware technology has developed so much as to allow for hugely more “intelligent” outcomes than were possible back then.

One example? Babel Fish / SYSTRAN worked mainly based on analytical gammar, i.e. the same system as the one [we think] we use for the job: but Translator is based mainly on statistical pattern matching.

Compared to a human professional translator, Google Translator’s performance is naive. But compared to >95% of humans, it is superior.

Enough said, I guess. Just sit back and watch the singularity unfold.