Bad Robot, Good Robot

Blog post from Dec 05, 2017

The advent of self-learning artificial intelligence (AI) in our daily lives is also accompanied by a growing fear of it. There are so many examples of failures and sentient machine takeover scenarios. On the one side, there are funny aspects such as the chatbot of a German fashion and footwear dealer, which advised a male writer to wear high-heeled sandals.

On the other side, there are more sobering aspects such as the fictional scenario of the German satire show "Extra 3". In the scenario, a family first gets a virtual assistant for their home but then tries to get rid of it when it starts acting independently. However, the plan fails because the AI has meanwhile transferred the money to a numbered account in the Bahamas.

The "virtual colleagues" have also started getting active in the field of texts. But not every software is a "bad robot".

Machine Assistance

It would not be good to trust text robots blindly, as they can only "produce" standardized content such as TV tips, financial information, and sports reports. This is not always successful and can even be dangerous, e.g. if enterprises who are subject to competitive pressure believe the claim that text bots can automatically generate SEO-relevant content in countless languages. In reality, such procedures must first master the algorithms of market giants like Google, which identify low-quality content and punish it with an inferior ranking. Still, machines can help people to create engaging, surprising, or even emotional texts. Our linguistic author assistance tool proves that this is actually possible.

The software automatically processes language. It is available at all times and accurately checks the spelling, grammar, and style in fractions of seconds. The text analysis begins at the character level. The software examines which characters are language-related, which ones not, and where the word boundaries are. On this basis, it conducts a linguistic analysis of the word components. It detects sentence boundaries and sentence elements, usually on the basis of punctuation marks and uppercase and lowercase spelling. Subsequently, the sentence-based text processing examines the individual sentences, looking for the necessary components for the respective sentence type and thus identifying errors in the sentence context. It can even improve its skills by adding new words. Does it sound difficult? Well, it's not!

Benefits of Augmented Intelligence

The writers, authors, reviewers, and decision-makers always remain in control. They decide when to use the author assistance. For example, they can use it as a real-time check that immediately analyzes the text written in the editor and submits improvement proposals. Or they can use the batch check as a machine review before forwarding the text. As a server-side process, the author assistance tool analyzes all new texts, serves as a quality gateway that blocks the release if certain thresholds are breached, and sends improvement proposals to the author. The author can then apply the proposals with a mouse click, insert them manually, or discard them. The author assistance tool works like an augmented intelligence assistant.

The benefits of such a good robot include correct spelling and grammar – including the punctuation marks, spaces, gender, number, case, and subject-verb congruence. The software helps to watch out for reasonable word and sentence lengths, use of the active voice, positive wording, avoidance of buzzwords, certain writing styles, and consistent terminology. At the same time, the author can compose individual texts. This helps enterprises to align their texts with their brand and objectives. Of course, the billing is transparent: We do not debit any amounts automatically, but we send you a fair invoice.

tl;dr

AI in content creation is not about "bad robots" which make authors superfluous. Rather, authors can stay in control while using AI technologies for authoring assistance to optimize their content.