NTT DOCOMO has introduced AI EQ+1 Framework to assess conversational AI beyond response speed and basic safety. Its reference architecture adds components for interpreting context, choosing behavior and improving conversations around an existing language model. Four supporting technologies address user-state assessment, pauses in speech, empathy with safety and learning from conversation logs.
NTT DOCOMO, the Japanese telecommunications company, announced AI EQ+1 Framework on October 26. It assesses how conversational AI behaves toward a person in a particular situation. Alongside speed, safety and honesty, the framework examines empathy, adaptation to context and consistency of character. It also considers outcomes such as attachment to a service and willingness to keep using it.[1], [2]
The accompanying EQ reference architecture describes components around an existing language model. They interpret context, choose behavior, express responses and guard against inappropriate answers, then support improvements to later conversations. DOCOMO says the components can be added without rebuilding the central model. The design separates those supporting functions from the model that supplies the underlying language capability.[1], [2]
Four supporting technologies generate user-state assessment rules, reduce pauses between text generation and speech, combine empathy with safety, and use conversation logs to improve responses. DOCOMO published a whitepaper covering the framework and experiments conducted between August and October. The company identifies customer support, care, education and entertainment as intended applications. Claims about emotional safety and continued use remain company-reported findings.[1], [2]
