Deepfakes: what did Obama just say?

17

September

2019

5/5 (1)

download-2-e1556701896433

It is getting easier to make a “deepfake video”. In the beginning of this month, an application in China went viral through which anyone is able to put his or her head in a movie scene of your choice. For this application, a Chinese phone number is required. What might be the effects when this technology would be easily accessible to everyone with a smartphone?

The technology making deepfakes possible is already existing for years. However, it began to get  popular when it was firstly used for porno movies. The use of deepfakes for porn is questionable, since the concerned person has mostly not given their approval. Therefore, Rebbit, the forum where it all started, has banned the controversial deepfake videos. Imagine that once the technology is easily accessible to everyone, even minors might create virtual porn, which is punishable in most countries. 

The public prosecutor of the Netherlands is worried about the use of the fake videos to persuade people to do something punishable for example. As a result, the maker of the concerned deepfake video can be prosecuted for defamation or slander in most countries. 

Besides that, a deepfake video could be used for blackmail, which could lead to dangerous situations. Researchers argue that in six to twelve months from now, you are probably not capable of distinguish a deepfake from a real video (NOS, 2019). One of the reasons is that it is currently complicated to falsify audio. 

The question is whether prohibitions are the right way to prevent the dangers of deepfakes, since social media is also not banned. It is not about the technique, but about what you do with the deepfakes (NOS, 2019). For example, it might be utilized for educational as well as psychological purposes. Think about virtually talking to a deceased loved one or teach a class with virtual historical figures. Regulatory institutions should reconsider the law to make these applications possible. 

Sources

Nos.nl (2019). Nederlandse Publieke Omroep: Zorgen OM over deepfakes: ‘Risico op oplichting en afpersing’. [online] Available at: https://nos.nl/artikel/2300688-zorgen-om-over-deepfakes-risico-op-oplichting-en-afpersing.html [Accesed 17 Sep. 2019]. 

Please rate this

Machine Learning to redesign hospital alarms and reduce hospital noise

9

September

2019

No ratings yet.

Imagine you are working as a nurse at the intensive care department in a hospital. Statistics show you will hear 150 to 200 hospital alarms a day, of which approximately 90% is false. This creates a lot of unnecessary and avoidable amount of stress. Moreover, there might be a risk that legit hospital alarms would not stand out among the false hospital alarms. This also leads to stressful situations for patients and their families. Hospital alarms can also disrupt the development of premature babies. Although it is hard to imagine modern health care without these hospital alarms, they might be destructive in the situations as described above. Existing hospital alarms no longer provide sufficient safety.

One of the challenges is the amount of data in a electronic health record (EHR). Traditional models choose a limited number of variables, resulting in – either negative or positive – false hospital alarms. An effective alarm merely goes off when a serious complication evolves, a doctor recognizes the alarm as symptomatic for the said issue and the required expertise to address the issue at hand exists.

The Máxima Medisch Centrum and the Technical University in Eindhoven, the Netherland, are cooperating to tackle these problems. Algorithms can help ringing more useful hospital alarms. Data of sensors of someone’s breathing and heartbeat can be combined, which results in an alarm that rings 20 seconds earlier when detecting an irregularity of the breathing and the heartbeat. Through Machine Learning, relations and patterns can be recognized which would not be recognized by humans. For example, mathematical models are able to detect potentially dangerous movements of the premature babies in their incubators by linking these movements to previous situations. This type of Artificial Intelligence is able to learn “baselines” and consolidates real-time trends to process alarms. As a result, currently disintegrated hospital alarms could transform into a more undivided, patient-centered analytic monitoring model. In this modern approach, the core transfers from individual hospital alarms to an integrated set of data from all patient-related equipment.

Some of these solutions can already be implemented, while others would take years before they become beneficial. This can be caused by strict regulations for medical equipment. However, Machine Learning could really help to prevent the negative consequences of a hospitalization.

Sources:

Rajkomar, A., Oren, E., Chen, K., Dai, A., Hajaj, N., Hardt, M., Liu, P., Liu, X., Marcus, J., Sun, M., Sundberg, P., Yee, H., Zhang, K., Zhang, Y., Flores, G., Duggan, G., Irvine, J., Le, Q., Litsch, K., Mossin, A., Tansuwan, J., Wang, D., Wexler, J., Wilson, J., Ludwig, D., Volchenboum, S., Chou, K., Pearson, M., Madabushi, S., Shah, N., Butte, A., Howell, M., Cui, C., Corrado, G. and Dean, J. (2018). Scalable and accurate deep learning with electronic health records. NPJ Digital Medicine, 1(1).

Chopra, V. and McMahon, L. (2014). Redesigning Hospital Alarms for Patient Safety. JAMA, 311(12), p.1199.

Nederlandse Omroep Stichting. Slimme algoritmes moeten ‘alarmstress’ in ziekenhuis voorkomen. Derived from:
https://nos.nl/artikel/2298745-slimme-algoritmes-moeten-alarmstress-in-ziekenhuis-voorkomen.html

Please rate this