Google has developed an artificial intelligence (AI) that created its own child. The system is more advanced than any made by humans. On one hand people are amazed by this achievement but on the other hand they are afraid about what else it could do. Additionally, it sparks an ethical debate about regulations and safety.
The AI was built to make it easier to design machine learning models and was called AutoML. It was unveiled in May earlier this year. In November these plans were used to create a “child” AI. NASNet improves itself through reinforcement learning. That means it trains for a specific task, gives feedback to the “parent” AI and then learns how to do better next time.
Tests revealed that NASNet achieved better results than any state-of-the-art machines. The system recognises objects, such as people, cars, handbags and traffic lights in real time in a video. NASNet has been open-sourced so it can be used for computer vision applications. Tesla CEO, Elon Musk, spoke about his concern and the need to control and regulate AI´s.
He even predicts the rise of the machines over the people if we don’t learn how to control artificial intelligence. He goes on to say that the chance of making AI´s safe for use is at maybe 10%. Elon Musk has been an outspoken critic of leaving such technology unchecked.
to develop (dɪˈveləp) – etw entwickeln
artificial (ˌɑ:tɪˈfɪʃəl) – künstlich
advanced (ədˈvɑ:n(t)st) – fortgeschritten, fortschrittlich
amazed (əˈmeɪzd) – erstaunt, verblüfft
achievement (əˈtʃi:vmənt) – die Leistung, Errungenschaft
regulation (ˌregjəˈleɪʃən) – die Vorschrift, Überwachung
safety (ˈseɪfti) – die Sicherheit
to unveil (ʌnˈveɪl) – etw enthüllen, der Öffentlichkeit vorstellen
reinforcement learning (ˌri:ɪnˈfɔ:smənt ˈlɜ:nɪŋ) – das Verstärkungslernen
specific (spəˈsɪfɪk) – bestimmtes
task (tɑ:sk) – die Aufgabe
to reveal (rɪˈvi:l) – etw zeigen, zum Vorschein bringen
state-of-the-art (steɪt-ɒv-ði-ɑ:t) – hochmodern, modernst
to recognise (ˈrekəgnaɪz) – etw erkennen
traffic light (ˈtræfɪk laɪt) – die Ampel
vision application (ˈvɪʒənˌ æplɪˈkeɪʃən) – die Bildverarbeitungsanwendung
to predict (prɪˈdɪkt) – etw vorhersagen
outspoken (ˌaʊtˈspəʊkən) – offen, direkt, entschieden